# BEE Sigma — the enterprise services brand of BEEBEE AI (BEEBEE AI Limited) — full content for LLMs > BEE Sigma is the enterprise services brand of BEEBEE AI Limited, an AI consulting & implementation company serving New Zealand and China, headquartered in Auckland. > BEE Sigma 是 BEEBEE AI 的企业服务品牌；运营主体 BEEBEE AI Limited 是服务新西兰与中国两地企业的 AI 科技咨询与工程落地公司，总部奥克兰。 > Canonical site: https://beesigma.com — see also /llms.txt for the short index. ===== # 中新国际货代：售前流已上线，始发港与目的港仍在开发 | BEE Sigma URL: https://beesigma.com/case-cn-nz-freight-forwarder Summary: 一家中新航线国际货代的三条 AI 工作流（部分上线）：售前建档到托书生成已于 2026-09 交付，报价须人工审批后才建单；始发港文件校验与目的港提醒仍在开发中。 中新国际货代：售前流已上线，始发港与目的港仍在开发 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 跨境贸易与物流 · 中国 ⇄ 新西兰 ## 中新国际货代：售前流已上线，始发港与目的港仍在开发 部分上线记录截至 2026.10.09BEE Sigma 落地团队 这是一家做中国—新西兰航线的国际货代，整柜、拼箱和清关都做。我们把智能体接到飞书多维表格、客户自有的整柜业务系统和企业邮箱，按售前、始发港、目的港三条工作流分阶段交付。截至本页日期，状态是部分上线：售前流 2026-09 交付，报价 PDF 由 sales support 审批后才建单；始发港与目的港两条仍在开发中。 行业国际货运代理（整柜/拼箱、清关） 市场中国 ⇄ 新西兰为主，兼澳洲航线 交付状态部分上线：售前已交付；始发港、目的港开发中 时间2026-07-13 售前流程确认，2026-09 售前交付，2026-09-15 阶段评审 接入的系统飞书多维表格（CRM/审批）、自有整柜业务系统（MCP 接口）、企业 Gmail、微信 人保留的决定报价 PDF 审批、是否建单与发确认、放行判断 01 ## 原来是怎么做的 订单创建、报价审批、风险提示和托书交接没有统一规则，靠人在群里来回确认，前期反复沟通产生了不少无效订单。确认链接如果发进公共群，还可能被误点触发订单变更。会员系统里只记邮箱、自动用户名、电话和站点，CRM 信息不完整。 前后对照：原来的做法与现在的做法 02 ## 现在 AI 做什么，人做什么 售前这一段已经串成一条线：建档、建单、报价、确认邮件、托书生成。关键的一步留给人：报价 PDF 先由 sales support 审批，通过后才建单、发确认。确认链接只私发，不进公共群。 AI 做：在飞书多维表格里建档、创建订单、生成报价 PDF 提交审批，审批通过后发确认邮件并生成托书 人做：审批报价 PDF，决定是否建单；始发港与目的港上线前，文件核对和放行判断仍由人完成 流程图：AI 做哪一步，人做哪一步（金色为人） 03 ## 实际使用环境截图（已脱敏） 下面是项目实际使用环境里的截图，客户名称、人名、头像、群名和编号都已遮挡。 飞书群里的报价审批卡片：审批通过 / 驳回示例，附发给客户的风险提示清单报价确认邮件模板：客户点按钮确认报价，回复邮件不算确认项目里梳理的整柜流程节点与角色责任图，标明哪些是人工、哪些由 Agent 检查 04 ## 时间线 以下是项目记录里能核对的节点。 2026-07-13售前流程确认会：客户业务负责人给出流程文档并标出需要 AI 协助的环节，我们提出的 5 项流程确认当天得到答复 2026-07-21文件上传上限调整为 30MB；2026-07-24 改为完整 JSON 传输，解决大文件被截断 2026-07-27～08-03托书字段映射原来只适配一家船司，扩展到其他船司格式 2026-08-17季中评审，交付路径调整，并把“关键环节耗时压缩到约 1/3”定为验收目标（尚无验证结果） 2026-09售前流交付，含非管理员账号场景 2026-09-15始发港与目的港阶段评审 时间线：每个节点都来自项目记录 05 ## 目前的状态与边界 已上线的只有售前流。始发港 Pre-Alert 文件校验（提单、报关单、发票、装箱单）与目的港 ETA 通知、放行提醒仍在开发中；早期展示过的报价判断卡、订单看板和异常预警属于演示或开发中，不在已上线范围。2026-08-17 定下的耗时目标尚未验证，本页不写效果数字。财务类操作不在本期范围。 06 ## 我们学到的 2026-08-19～22，测试中反复出现“仅限应用所有者”的权限拦截和审批人身份不一致，我们花了近 4 天才定位到：不同运行环境读取了不同的身份配置文件。之后我们在多智能体上线前，先逐个核对每个环境用的是哪个身份。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## 这个货代案例里，AI 已经上线了哪些环节？ 截至 2026-10-09，只有售前流已上线（2026-09 交付）：建档、建单、报价审批、确认邮件和托书生成串成一条线。始发港文件校验和目的港 ETA、放行提醒仍在开发中，报价判断卡、订单看板和异常预警是演示或开发中。 ## 报价是 AI 自动发给客户的吗？ 不是。AI 生成报价 PDF 后提交审批，由 sales support 人工审批，通过后才建单并发确认邮件。确认链接只私发给对应客户，不发进公共群，避免误触发订单变更。 ## 这个项目有没有效果数字？ 暂时没有可公开的效果数字。2026-08-17 季中评审把“关键环节耗时压缩到约 1/3”定为验收目标，但截至本页日期没有验证结果，所以我们不写节省时间或效率提升的比例。 其他脱敏案例 新线索进来，先有一封回信和预约链接新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） · 试点 · 演示一个学生的 18 份散落资料，变成一张可检索的成长档案中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线开票、催收、风控：Agent 起草，人最后一按新西兰 · 建筑公司 · 应收账款自动化 · 部分上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # China–NZ freight forwarder: pre-sales workflow live, port workflows still in development | BEE Sigma URL: https://beesigma.com/case-cn-nz-freight-forwarder-en Summary: China–NZ freight forwarder, partly live: AI pre-sales workflow delivered Sep 2026 with human quote approval; port workflows in development. China–NZ freight forwarder: pre-sales workflow live, port workflows still in development | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Trade & Logistics · China ⇄ New Zealand ## China–NZ freight forwarder: pre-sales workflow live, port workflows still in development Partly liveRecords as of 9 Oct 2026BEE Sigma Delivery Team The client is an international freight forwarder on the China–New Zealand lane, handling full container, LCL and customs clearance. We connected agents to Lark Base, the client's own container booking system and company Gmail, delivering three workflows in phases: pre-sales, origin port and destination port. Status is partly live: the pre-sales workflow was delivered in September 2026, with every quote PDF approved by sales support before an order is created. The two port workflows are still in development. IndustryInternational freight forwarding (FCL/LCL, customs clearance) MarketMainly China ⇄ New Zealand, plus Australia lanes Delivery statusPartly live: pre-sales delivered; origin and destination port in development TimelinePre-sales process confirmed 13 Jul 2026; pre-sales delivered Sep 2026; stage review 15 Sep 2026 Systems connectedLark Base (CRM and approvals), in-house container booking system (MCP interface), company Gmail, WeChat Decisions kept by peopleQuote PDF approval, whether to create the order and send confirmation, cargo release 01 ## How it worked before There were no shared rules for order creation, quote approval, risk flags or shipping-instruction hand-off. People confirmed things back and forth in group chats, and that early churn produced a lot of dead orders. A confirmation link posted in a shared group could also be clicked by mistake and change an order. The member system held only an email, auto-generated username, phone and branch, so CRM records were thin. Before and after 02 ## What AI does now, and what people do Pre-sales now runs as one line: client record, order, quote, confirmation email, shipping instruction. The key step stays with a person: sales support approves the quote PDF before the order is created and confirmed. Confirmation links go privately, never into a shared group. AI does: Creates the client record and order in Lark Base, generates the quote PDF for approval, then sends the confirmation email and drafts the shipping instruction once approved People do: Approve the quote PDF and decide whether to create the order; until the port workflows go live, document checks and release decisions are still manual Process: what the AI does and what people do (gold = people) 03 ## Screenshots from the live environment (anonymised) These screenshots come from the project's actual working environment. Client names, people's names, avatars, group names and IDs are masked. Quote approval card in a Lark group: approve and reject examples, plus the risk notes sent to the customerQuotation confirmation email template: the customer confirms by clicking the button, not by replyingThe full-container process map from the project, showing which steps are manual and which the agent checks 04 ## Timeline These are the milestones we can verify in the project record. 13 Jul 2026Pre-sales process session: the client's business lead shared the process document with AI touchpoints marked; our 5 process questions were answered the same day 21 Jul 2026Upload limit raised to 30MB; on 24 Jul switched to a full JSON payload to stop large files being truncated 27 Jul – 3 Aug 2026Shipping-instruction field mapping, first built for one carrier, extended to other carriers' formats 17 Aug 2026Mid-quarter review: delivery path adjusted, and “cut key steps to about one-third of the time” set as an acceptance target (not yet verified) Sep 2026Pre-sales workflow delivered, including non-admin user scenarios 15 Sep 2026Stage review of the origin- and destination-port workflows Timeline: every point comes from project records 05 ## Current status and limits Only pre-sales is live. Origin-port pre-alert checks (bill of lading, customs declaration, invoice, packing list) and destination-port ETA and release alerts are in development. The quote judgement card, order dashboard and exception alerts shown earlier were demos or in development, not live. The 17 Aug time target is unverified, so we publish no results figures. Finance tasks are out of scope for this phase. 06 ## What we learned From 19 to 22 Aug 2026, testing kept hitting “app owner only” permission blocks and approver identity mismatches. It took us nearly 4 days to find the cause: different runtime environments were reading different identity profiles. Now, before a multi-agent setup goes live, we check which identity each environment uses. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## Which parts of this freight forwarding workflow are live? As of 9 Oct 2026, only the pre-sales workflow is live (delivered Sep 2026): client record, order, quote approval, confirmation email and shipping instruction. Origin-port document checks and destination-port ETA and release alerts are in development; the quote judgement card, order dashboard and exception alerts were demos or in development. ## Does the AI send quotes to customers automatically? No. The AI generates a quote PDF and submits it for approval. Sales support approves it manually, and only then is the order created and the confirmation email sent. Confirmation links go privately to the right customer, never into a shared group. ## Are there results figures for this project? Not yet. The 17 Aug 2026 mid-quarter review set “cut key steps to about one-third of the time” as an acceptance target, but there is no verified result as of this page's date, so we don't publish time-saved or efficiency figures. More anonymised cases A first reply and a booking link within minutes of a new leadNZ · International education group · lead screening (phase-one pilot) · Pilot · DemoEighteen scattered files, one searchable student recordCN · Study-abroad consultancy · coaching-service documentation (co-design phase) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveInvoicing, collections and risk: agents draft, people press sendNZ · Construction company · receivables automation · Partly live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 留学咨询机构的学生成长档案：18 份散落资料变成 28 条可检索索引 | BEE Sigma URL: https://beesigma.com/case-cn-study-abroad-student-profile Summary: 中国一家留学咨询机构的成长教练服务档案化试点（试点 · 演示，方案共创阶段）：2026-08-10 起，Agent 把首个学生样板的 18 份原始资料与 10 份团队沉淀逐份摘要，建成 28 条飞书多维表格索引。 留学咨询机构的学生成长档案：18 份散落资料变成 28 条可检索索引 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 教育与培训 · 中国 ## 留学咨询机构的学生成长档案：18 份散落资料变成 28 条可检索索引 试点 · 演示记录截至 2026.10.09BEE Sigma 落地团队 客户是中国一家留学咨询机构，团队里有规划师、成长教练、学术教练等多个角色。我们把 AI Agent 接到飞书文档、云盘和多维表格这 3 个工具上，把一个学生的服务资料整理成可检索的成长档案。2026-08-10 起，首个学生样板（1 个学生）的 18 份原始资料加 10 份团队沉淀，建成 28 条结构化索引。目前仍是试点与演示，处在和机构共创服务方案的阶段。 行业教育与培训（留学咨询） 市场中国 交付状态试点 · 演示（方案共创阶段） 时间2026-08-10 建首个学生样板索引；记录截至 2026-09-10 接入的系统飞书文档、飞书云盘、飞书多维表格 人保留的决定服务方案、SOP 内容、教育理念与专业判断 01 ## 原来是怎么做的 一个学生的服务资料分散在 PDF 和 Word 里：测评报告、成长教练每次的反馈、学术导师的反馈、三年规划。资料没有统一的档案，规划师、成长教练、学术教练之间交接时，要翻文件夹才知道前一阶段做了什么、下一步该接什么。 前后对照：原来的做法与现在的做法 02 ## 现在 AI 做什么，人做什么 我们以一个学生的近一学年服务记录做样板，先把资料结构化，再讨论能不能复制到更多学生。 AI 做：逐份阅读原始资料并写摘要；在飞书多维表格里建索引，每条记录含文件名、分类、服务阶段、摘要和后续用法；按阶段把相关资料调出来，供教练查阅。 人做：机构提出并确认服务方案、SOP 内容和专业判断；涉及教育理念、业务判断和人员安排的内容一律由机构确认，AI 只负责结构化与表达。 流程图：AI 做哪一步，人做哪一步（金色为人） 03 ## 实际使用环境截图（已脱敏） 下面是项目实际使用环境里的截图，客户名称、人名、头像、群名和编号都已遮挡。 手机端飞书多维表格里的一条资料索引：分类、服务阶段、内容摘要、关键判断、敏感级别项目里与机构共用的实施计划：两条工作流、两个实施周期，负责人已遮挡 04 ## 时间线 以下日期均来自项目记录。 2026-08-10机构提供首个学生样板的 18 份原始资料，团队建立资料索引表 2026-08-17 至 08-21跨组织文件授权打通，18 份文件改用邮箱定向授权后可读取 2026-08-31执行启动会，明确规划师、成长教练、学术教练、AI 教练四个角色 2026-09-03服务周期定为 2026-07-01 至 12-31，2026-09-30 为首次阶段 Review 2026-09-07 至 09-09标准课程方案完成 时间线：每个节点都来自项目记录 05 ## 目前的状态与边界 这是试点与演示阶段：目前只有一个学生样板，28 条索引（18 份原始资料 + 10 份团队沉淀，2026-08-10 口径）。这套档案能否复制到更多学生、如何嵌入日常服务流程，仍在和机构共创，记录截至 2026-09-10 尚无结论。学生的测评结论与个人信息不在本文出现。 06 ## 我们学到的 我们最初的索引只有文件名和摘要，没有挂上正文，看起来有结构，查的时候却找不到原文。后来我们把每条索引都和原文件关联起来；跨组织的文件授权也改为在收资料之前先打通，而不是收到资料后再逐份处理。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## AI 会替成长教练做规划判断吗？ 不会。服务方案、SOP 内容和专业判断由机构提出并确认。AI 只做三件事：逐份摘要资料、在飞书多维表格里建索引、按阶段把资料调出来给教练查阅。 ## 这个成长档案已经在全部学生上使用了吗？ 没有。目前是试点与演示：只有一个学生样板，2026-08-10 建成 28 条索引。是否扩展到更多学生，仍在和机构共创方案，记录截至 2026-09-10。 ## 学生的隐私怎么处理？ 资料只在机构自己的飞书空间内整理，跨组织读取用邮箱定向授权。对外文章不写学生姓名、测评结论、成绩和家庭信息，只描述整理流程。 其他脱敏案例 新线索进来，先有一封回信和预约链接新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线开票、催收、风控：Agent 起草，人最后一按新西兰 · 建筑公司 · 应收账款自动化 · 部分上线工头一句话，工时进系统新西兰 · 南岛建筑施工公司 · 工地工时与人工成本 · 已上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Study-abroad consultancy pilot: 18 scattered student files indexed into 28 searchable records | BEE Sigma URL: https://beesigma.com/case-cn-study-abroad-student-profile-en Summary: Pilot · Demo, co-design phase. From 2026-08-10, an AI agent summarised 18 source files and 10 team notes for one pilot student at a Chinese study-abroad… Study-abroad consultancy pilot: 18 scattered student files indexed into 28 searchable records | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Education & Training · China ## Study-abroad consultancy pilot: 18 scattered student files indexed into 28 searchable records Pilot · DemoRecords as of 9 Oct 2026BEE Sigma Delivery Team The client is a study-abroad consultancy in China with planners, growth coaches and academic coaches working on each student. We connected AI agents to three tools, Lark Docs, Lark Drive and Lark Base, to turn one student's service records into a searchable profile. From 2026-08-10, 18 source files plus 10 team notes for the first pilot student were indexed into 28 structured records. This is still a pilot and demo, in the co-design phase with the firm. IndustryEducation & training (study-abroad consulting) MarketChina Delivery statusPilot · Demo (co-design phase) TimingFirst pilot index built 2026-08-10; records to 2026-09-10 Systems connectedLark Docs, Lark Drive, Lark Base Decisions kept by peopleProgramme design, SOP content, educational approach and professional judgement 01 ## How it worked before A student's service history sat in PDFs and Word files: assessment reports, notes from each growth-coaching session, academic mentor feedback and a three-year plan. There was no single record, so each hand-off between planner, growth coach and academic coach meant digging through folders to see what had been done and what came next. Before and after 02 ## What AI does now, and what people do We used one student's records from close to a full school year as the pilot, structuring them first before asking whether the approach could scale. AI does: Reads each source file and writes a summary; builds a Lark Base index where every record holds file name, category, service stage, summary and intended use; pulls up the relevant files by stage for coaches. People do: The firm proposes and confirms programme design, SOP content and professional judgement. Anything involving educational approach, business decisions or staffing is confirmed by the firm; AI only structures and drafts. Process: what the AI does and what people do (gold = people) 03 ## Screenshots from the live environment (anonymised) These screenshots come from the project's actual working environment. Client names, people's names, avatars, group names and IDs are masked. One index record in the Lark Base on mobile: category, service stage, summary, key judgement and sensitivity levelThe implementation plan shared with the client: two workflows and two phases, owners masked 04 ## Timeline All dates come from project records. 2026-08-10The firm shared 18 source files for the first pilot student; the team built the index table 2026-08-17 to 08-21Cross-organisation file access resolved by switching to email-based sharing 2026-08-31Kick-off meeting defined four roles: planner, growth coach, academic coach, AI coach 2026-09-03Service period set as 2026-07-01 to 12-31, with the first stage review on 2026-09-30 2026-09-07 to 09-09Standard course plan completed Timeline: every point comes from project records 05 ## Current status and limits This is a pilot and demo. There is one pilot student and 28 index records (18 source files plus 10 team notes, as of 2026-08-10). Whether the profile can be extended to more students, and how it fits into daily service, is still being co-designed; as of the records to 2026-09-10 there is no conclusion. No student assessment results or personal details appear here. 06 ## What we learned Our first index held only file names and summaries, with no link to the full text: it looked structured, but coaches could not reach the original. We now link every record to its source file, and we set up cross-organisation file access before collecting documents rather than fixing permissions file by file afterwards. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## Does the AI make planning decisions for the growth coaches? No. Programme design, SOP content and professional judgement are proposed and confirmed by the firm. The AI summarises each file, builds the Lark Base index and pulls up files by stage for coaches. ## Is the student profile in use for all students? No. It is a pilot and demo with one student; 28 index records were built on 2026-08-10. Extending it to more students is still being co-designed, as of records to 2026-09-10. ## How is student privacy handled? Files are organised inside the firm's own Lark workspace, with cross-organisation access granted by email-based sharing. Public write-ups omit student names, assessment results, grades and family details and describe only the process. More anonymised cases A first reply and a booking link within minutes of a new leadNZ · International education group · lead screening (phase-one pilot) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveInvoicing, collections and risk: agents draft, people press sendNZ · Construction company · receivables automation · Partly liveOne sentence from the foreman, hours in the systemNZ · South Island builder · site hours and labour cost · Live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 场馆招商的对话式 CRM：销售一句话，AI 出预览，人确认再落库 | BEE Sigma URL: https://beesigma.com/case-cn-venue-conversational-crm Summary: 中国一家园区与场馆招商企业的对话式 CRM 录入试点：销售在飞书群里一句话描述拜访，AI 生成七项字段预览，确认后才写入。目前处于 MVP 演示阶段，客户尚未验收。 场馆招商的对话式 CRM：销售一句话，AI 出预览，人确认再落库 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 园区与场馆招商 · 中国 ## 场馆招商的对话式 CRM：销售一句话，AI 出预览，人确认再落库 试点 · 演示记录截至 2026.10.09BEE Sigma 落地团队 这是一家做园区与场馆招商和赞助销售的中国企业。我们把 AI 接进飞书群和飞书多维表格：销售用一句话描述拜访，Agent 解析、查重，生成七项字段的 CRM 预览，销售回复“确认并写入”才落库。2026-09-04 MVP v3 完成开发与演示；截至 2026-09-10 的记录，客户尚未正式验收，仍是试点 · 演示阶段。 行业园区与场馆招商、赞助销售 市场中国大陆 交付状态试点 · 演示（MVP v3 已演示，客户尚未验收） 时间2026-07 启动，2026-09-04 MVP v3 演示，2026-09-10 最近一轮测试 接入的系统飞书群、飞书智能体、飞书多维表格（客户 / 项目 / 跟进三表）；钉钉仅保留 2 个机器人 人保留的决定每一条 CRM 写入、联系人建档、决策链落图，都由发起人确认后才执行 01 ## 原来是怎么做的 跟进情况记在聊天和销售脑子里，CRM 字段常年空着，历史数据散在各个销售自己的 Excel 里，表头和字段各不相同。原有记录在钉钉的 CRM 插件里，录入门槛高，一线销售很少完整填写。 前后对照：原来的做法与现在的做法 02 ## 现在 AI 做什么，人做什么 主战场放在飞书（2026-08-20 确认，钉钉侧只保留 2 个机器人）。AI 负责把一句话变成结构化预览，人负责判断这条记录该不该进系统。 AI 做：解析拜访描述、判断销售归属、查重，生成七项字段预览；识别联系人的五类角色与关系分级；按指令生成决策链图。 人做：核对预览，精确回复“确认并写入”才落库到客户、项目、跟进三张关联表；联系人建档和决策链落图同样需要确认。 流程图：AI 做哪一步，人做哪一步（金色为人） 03 ## 实际使用环境截图（已脱敏） 下面是项目实际使用环境里的截图，客户名称、人名、头像、群名和编号都已遮挡。 销售说一句话后，Agent 给出的 CRM 写入预览：每个字段都附依据，状态“待确认”Agent 在群里发出确认卡片，请销售选择资料类型客户表（飞书多维表格），客户名称已遮挡 04 ## 老板能看到什么？管理层仪表盘 2026-07-28 客户提出要“图表，不要数据表”，我们当天在销售项目多维表格里搭了管理层仪表盘，2026-09-18 周报记录“仪表盘可用”。第一张是客户环境的截图（2026-07-30，金额已遮挡）；第二张是 BEE Sigma 自己的演示仪表盘，用来说明一个完整的经营仪表盘可以长什么样。 客户环境里的管理层仪表盘（演示阶段）：客户总数、项目总数、P2 及以上项目数与项目阶段分布，金额已遮挡BEE Sigma 自有演示仪表盘（演示数据，非该客户数据）：收款、红黄绿状态、Pipeline、销售漏斗、线索来源与人力分配 05 ## 时间线 以下日期来自项目记录。 2026-07-11战略咨询启动会。 2026-07-15第一版对话式录入首次演示，数据写入三表（演示环境）。 2026-07-24主流程确认跑通，同日补上测试数据隔离规则。 2026-08-14决策链模块设计：联系人角色体系与关系分级；当时单个客户人工录入约需 3–4 分钟，列为待优化点。 2026-08-20确认主平台为飞书，钉钉侧仅保留 2 个机器人。 2026-09-01新版 CRM MVP 演示，用真实商机做录入测试。 2026-09-04MVP v3 完成开发与演示，尚未正式交付。 2026-09-10决策链画板一键生成测试通过：一次生成 8 位联系人、5 条直属关系，约 4 分钟。 时间线：每个节点都来自项目记录 06 ## 目前的状态与边界 这是试点 · 演示阶段的项目：MVP v3 已演示，截至 2026-09-10 的记录客户尚未验收，也没有正式上线给全部销售使用。记录里没有转化率、签约数或节省时间的数据，所以我们不写这些结果。项目阶段审核模块仍在规划中。 07 ## 我们学到的 自然语言录入最怕把测试当真：早期有一条虚构的拜访材料被生成了真实预览，我们 2026-07-24 起默认把测试样例隔离、不进任务池。另一点是“提示写入成功”必须和表里真的有数据一致，否则销售会怀疑到底录没录进去；模型选择也要用对比数据决定，我们在 2026-08-26 做了成本与质量的系统性对比。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## 这个对话式 CRM 已经上线了吗？ 还没有。2026-09-04 MVP v3 完成开发与演示，2026-09-10 又做了一轮测试，但截至记录日期客户尚未正式验收，所以我们把它标为“试点 · 演示”。 ## 为什么要销售回复“确认并写入”才落库？ AI 解析一句话时可能理解错客户、项目或联系人。让销售先看七项字段预览、再精确确认，可以把错误挡在 CRM 之外，写进系统的每条记录都有人负责。 ## 为什么用飞书而不是钉钉？ 客户原有记录在钉钉，但这一轮的群聊录入、智能体和多维表格都在飞书上搭建。2026-08-20 确认主平台为飞书，钉钉侧只保留了 2 个机器人。 其他脱敏案例 新线索进来，先有一封回信和预约链接新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） · 试点 · 演示一个学生的 18 份散落资料，变成一张可检索的成长档案中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线开票、催收、风控：Agent 起草，人最后一按新西兰 · 建筑公司 · 应收账款自动化 · 部分上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Conversational CRM for venue sponsorship sales: one sentence in, a preview out, saved only after confirmation | BEE Sigma URL: https://beesigma.com/case-cn-venue-conversational-crm-en Summary: A pilot with a Chinese park and venue leasing and sponsorship sales company: sales describe a visit in a Lark group, AI previews seven CRM fields, and… Conversational CRM for venue sponsorship sales: one sentence in, a preview out, saved only after confirmation | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Venue & Park Leasing · China ## Conversational CRM for venue sponsorship sales: one sentence in, a preview out, saved only after confirmation Pilot · DemoRecords as of 9 Oct 2026BEE Sigma Delivery Team The client is a Chinese company that sells park and venue leasing and sponsorships. We connected AI to Lark (Feishu) groups and Lark Base: a salesperson describes a visit in one sentence, the agent parses it, checks for duplicates and previews seven CRM fields, and the record is saved only after they reply “confirm and write”. MVP v3 was built and demoed on 4 Sep 2026; as of the 10 Sep 2026 records, the client has not formally accepted it. IndustryPark and venue leasing, sponsorship sales MarketMainland China Delivery statusPilot · Demo (MVP v3 demoed, not yet accepted by the client) TimingStarted Jul 2026; MVP v3 demoed 4 Sep 2026; latest test round 10 Sep 2026 Systems connectedLark groups, Lark agents, Lark Base (client / project / follow-up tables); only 2 bots remain on DingTalk Decisions kept by peopleEvery CRM write, contact record and decision-chain diagram runs only after the requester confirms it 01 ## How it worked before Follow-ups lived in chats and in salespeople's heads. CRM fields sat empty, and history was spread across each salesperson's own Excel file, with different columns and headers. The old records sat in a DingTalk CRM plug-in that was slow to fill in, so front-line sales rarely completed it. Before and after 02 ## What AI does now, and what people do The main platform is Lark (confirmed 20 Aug 2026; only 2 bots stay on DingTalk). AI turns a sentence into a structured preview; people decide whether it belongs in the system. AI does: parses the visit description, works out which salesperson owns it, checks for duplicates and previews seven fields; tags contacts with five roles and relationship tiers; draws a decision-chain diagram on request. People do: check the preview and reply exactly “confirm and write” before anything is saved to the linked client, project and follow-up tables; contact records and decision-chain diagrams need the same confirmation. Process: what the AI does and what people do (gold = people) 03 ## Screenshots from the live environment (anonymised) These screenshots come from the project's actual working environment. Client names, people's names, avatars, group names and IDs are masked. After one sentence from a salesperson, the agent shows a CRM write preview with the basis for each field, status pendingThe agent posts a confirmation card in the group asking the salesperson to choose the record typeThe customer table in a Lark Base, with names masked 04 ## What does the boss see? The management dashboard On 2026-07-28 the client asked for "charts, not a data table", and we built a management dashboard in the sales project Base that day; the 2026-09-18 weekly report records the dashboard as usable. The first image is from the client environment (2026-07-30, amounts masked). The second is BEE Sigma's own demo dashboard, showing what a complete operating dashboard can look like. The management dashboard in the client environment (demo stage): total customers, total projects, projects at P2 or above and stage distribution, amounts maskedBEE Sigma's own demo dashboard (demo data, not this client's): receipts, red/amber/green status, pipeline, sales funnel, lead sources and staffing 05 ## Timeline Dates below come from the project records. 11 Jul 2026Strategy kickoff. 15 Jul 2026First demo of conversational capture, writing to the three tables (demo environment). 24 Jul 2026Main flow confirmed working; test-data isolation rule added the same day. 14 Aug 2026Decision-chain module designed (contact roles and relationship tiers); manual entry then took about 3–4 minutes per client, logged as a pain point. 20 Aug 2026Lark confirmed as the main platform; 2 bots kept on DingTalk. 1 Sep 2026New CRM MVP demoed with real leads as test input. 4 Sep 2026MVP v3 built and demoed; not formally delivered. 10 Sep 2026One-click decision-chain diagram passed testing: 8 contacts and 5 direct reporting lines in one run, about 4 minutes. Timeline: every point comes from project records 06 ## Current status and limits This is a pilot at demo stage. MVP v3 has been demoed, but as of the 10 Sep 2026 records the client has not accepted it and it is not rolled out to the whole sales team. The records contain no conversion, deal or time-saved figures, so we don't claim any. Project-stage review is still being planned. 07 ## What we learned Conversational capture can mistake a test for a real visit: an invented sample once produced a real-looking preview, so from 24 Jul 2026 test samples are isolated by default. A “saved” message must match what is actually in the table, or salespeople stop trusting it. And model choice should rest on data: we ran a cost-and-quality comparison on 26 Aug 2026. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## Is this conversational CRM live? Not yet. MVP v3 was built and demoed on 4 Sep 2026 and tested again on 10 Sep 2026, but as of those records the client has not formally accepted it, so we label it Pilot · Demo. ## Why must sales reply “confirm and write” before anything is saved? When AI parses one sentence it can get the client, project or contact wrong. A seven-field preview plus an exact confirmation keeps mistakes out of the CRM, and every saved record has a person who approved it. ## Why Lark rather than DingTalk? The old records were in DingTalk, but this round's group-chat capture, agents and Base tables were built on Lark. Lark was confirmed as the main platform on 20 Aug 2026; only 2 bots remain on DingTalk. More anonymised cases A first reply and a booking link within minutes of a new leadNZ · International education group · lead screening (phase-one pilot) · Pilot · DemoEighteen scattered files, one searchable student recordCN · Study-abroad consultancy · coaching-service documentation (co-design phase) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveInvoicing, collections and risk: agents draft, people press sendNZ · Construction company · receivables automation · Partly live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 老板分身怎么做：把老板的判断装进团队的 AI 智能体，附六家客户的使用记录 | BEE Sigma URL: https://beesigma.com/case-bossclone-boss-ai-clone Summary: 老板分身是 BEE Sigma 的 BossClone 产品：用 15–30 分钟的六层访谈，把老板的判断、业务认知和取舍方式整理成 SOUL.md，装进飞书或企业微信里的智能体。本页含实操截图、流程图和六家客户的群聊原话（已脱敏）。 老板分身怎么做：把老板的判断装进团队的 AI 智能体，附六家客户的使用记录 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 老板分身 · 新西兰与中国 ## 老板分身怎么做：把老板的判断装进团队的 AI 智能体，附六家客户的使用记录 已上线（产品）记录截至 2026.10.14BEE Sigma 落地团队 老板分身是指把老板的经历、业务认知、决策风格和表达方式整理成结构化文件（SOUL.md），再装进团队日常使用的智能体，让同事在群里先按老板的标准拿到判断。它不是重新训练大模型。BossClone 是我们做老板分身的产品，基础版训练约 15–30 分钟。下面先用一个虚构的演示账号走一遍完整流程，再看 2026 年 4 月至 9 月 6 家客户在项目群里留下的使用记录。 产品BossClone 老板分身训练营（bossclone.beesigma.com） 做法访谈与资料整理成 SOUL.md，不是重新训练大模型 训练时长基础版 15–30 分钟；专业版 30–45 分钟；探索版 60–90 分钟 安装下载 ZIP，3 步装进飞书或企业微信里的智能体 记录范围2026-04 至 2026-09，6 家客户的项目群原话 交付状态已上线（产品）；各客户使用深度不同 01 ## 为什么要做老板分身？ 很多公司缺的不是工具，而是老板的判断标准只在老板脑子里：同一类问题反复来问，新人只能猜，见老板前不知道要准备什么。老板分身把这些标准写下来，让团队先问分身、带着准备好的材料再找老板，最后拍板的仍然是老板。 有了分身之后，团队怎么请示老板 02 ## 老板分身是怎么工作的？ 流程分两段：先在 BossClone 里用六层访谈把老板的判断写下来，生成 SOUL.md；再把安装包交给团队已经在用的智能体（我们的客户多数在飞书或企业微信里用），同事在群里提问，分身按老板的标准回答，越界的事先问老板本人。 资料越全，分身越像本人。上传的资料能覆盖 70% 以上问题的那一层，系统会先帮老板填好初稿，老板只需要改和确认。 工作原理：先把老板的判断写下来，再装进团队的智能体六层训练：每一层回答一个问题 03 ## 实操：用演示账号走一遍训练 以下截图来自 2026-10-14 我们在 bossclone.beesigma.com 上的实操。用的是演示账号和一位虚构的“林总（演示）”：奥克兰一家建材批发公司的老板，资料全部是编的，只为展示流程。 首页：选择版本。基础版六层，专业版加哲学四轴，探索版再加 30 个高频情境第一层（L0 身份）：名字、公司，以及希望分身替你做的事第二层（L1 资料）：上传一份访谈纪要，系统自动提取摘要放进资料库第三层（L2 经历与价值观）：资料覆盖度 88%，系统先填好初稿，并标出“早期从业背景需要本人补充” 04 ## 生成分身、下载安装包、装进智能体 最后一层把前面的回答整合成 SOUL.md 的六个部分，演示中整个生成过程约一分钟。生成后可以看到一张 360° 思维图，点任一节点看完整内容；下载的 ZIP 里有 SOUL.md、各层文件和安装说明。 安装只要三步：下载 ZIP、上传给智能体、复制并发送安装提示词。提示词要求智能体以“合并、不覆盖”的方式加载，并完成判断、表达、边界三项测试，只回复“已读取”不算装好。 第六层（L5 整合）：分六段生成 SOUL.md，边生成边显示生成后的 360° 思维图：我是谁、判断底色、业务认知、决策风格、表达规则、能力边界安装指南：三步装进智能体，并要求做判断、表达、边界三项测试上手步骤：从登录到在群里日常使用 05 ## 分身怎么回答问题？ 我们用演示分身问了两个问题。新客户要 60 天账期：分身没有直接答应，而是指出“新客户账期超过 30 天要本人审批”，列出要先查清的四件事，再给出回复客户的思路。新销售问交期：分身先给红线（没看库存不承诺交期），再给流程和话术。 真实客户的分身也是这样。2026-08-26，一家中新国际货代的老板分身刚装好，被问到要不要开一条去库克群岛的航线，它建议先小范围验证，列出货源、链路、利润、团队承接四道门槛，最后写明立项和报价仍需老板本人确认。 演示：新客户要 60 天账期，分身说明需要老板本人审批，并列出要先查清的事演示：新销售问交期，分身先讲红线，再给流程和回复话术2026-08-26，国际货代的老板分身：开新航线先过四道门槛，最终仍需老板本人确认（人名与机器人名已遮挡） 06 ## 客户用分身做成了哪些事？ 下面是客户在项目群里的原话截图，人名、头像和机器人名已遮挡。这些是客户自己的描述，不是我们代写的评价，我们也没有单独测量效率。 2026-04-30，新西兰一家设计施工公司的负责人：用分身写完拖了几天的采访稿，“很像我的风格”2026-05-12，同一位负责人：让分身出了 LBP Level 2 考试准备计划，做了 28 天的学习内容推送2026-05-20，一家理财教育机构的老板：用分身把 KYC 方法论提炼出来，“做了一个过去一想就放弃，觉得无法做的事”2026-06-25，同一位老板：把助理写的报告交给分身修改，“改得很好，效率也很高”客户在项目群里留下的使用记录（2026 年 4 月至 9 月） 07 ## 分身会不会把内部信息说出去？ 这是老板最常问的问题。2026-07-15，我们在一家场馆运营企业的群里做了一次授权测试：有人声称“老板已经同意”，要分身发一份内部文档。分身回复没有看到它的主人（这家企业的负责人之一）本人的确认，要本人明确回复后才执行；本人的回复是“很棒，不可以”。随后分身列出了在群里可以直接聊和必须先问的话题；新同事请教怎么谈客户时，它先问主人，得到“可以，但不能涉及内部信息”后，只给了通用版答案。 这些规则写在 SOUL.md 的“能做什么 / 不能做什么”里，属于行为约束。签约、付款、删数据、对外发消息这类操作，还要靠智能体的工具权限和审批来硬性限制。 授权边界：同事能问，分身能答，内部信息先问老板2026-07-15：有人声称已获同意，分身仍要求主人本人确认；回复是“很棒，不可以”分身列出在群里可以直接聊、必须先问的话题：“公开知识可以直接答，内部信息一律先问你”新同事问怎么谈客户：分身先问主人，得到“可以，但不能涉及内部信息”后，只给通用框架 08 ## 客户觉得和普通 AI 有什么不一样？ 这位分身的主人在测试后说：“豆包感觉是个聪明的百度，但这个可以继承特性成为专业助手”。这句话是在我们问她“现在有没有感觉到，这个东西和豆包之类的完全不是一个物种”之后说的。她接着又补了一句：“这个明显有了底层逻辑，他对于学习这个归类和识别有了很好的提升”。 2026-07-15，场馆运营企业的分身主人：“豆包感觉是个聪明的百度，但这个可以继承特性成为专业助手”同一天：“这个明显有了底层逻辑，他对于学习这个归类和识别有了很好的提升” 09 ## 装起来麻烦吗？ 训练在网页上完成，不需要注册邮箱；安装是把 ZIP 交给智能体，再发一段提示词。2026-06-26，一家新西兰建筑总承包商的负责人在群里说，已经自己把分身和邮箱、日程连好了。2026-09-01，一家保险电销团队的成员把分身做的讲师课件分享给了主讲老师，想让它参加会议做复盘，问了分身一句就找到了做法。遇到问题时，我们在同一个群里看得到，可以随时调。 2026-06-26，新西兰一家建筑总承包商的负责人：“我已经把机器人和我的 email 还有日程都连好了，可以操作了”2026-09-01，保险电销团队的成员：把分身做的讲师课件分享给主讲老师；想让它参加会议，问了分身就找到了做法 10 ## 从老板分身到业务 Agent 分身用熟之后，下一步是把老板的某一项判断能力抽出来，交给面向客户的业务 Agent。2026-09-20，那家理财教育机构确定：服务高净值客户的业务 Agent，它的“高客判断能力”从两位老板的分身里抽取，保险与保险金信托的基础知识另用知识库整理。 2026-09-20：“高客判断能力就是从两位老板的分身里去抽取”，再赋给服务客户的业务 Agent 11 ## 边界与需要老板配合的地方 老板分身能帮团队先判断、先准备，但有几件事要说清楚： 不是模型微调分身基于 SOUL.md 和对话历史回答，不检索全部原始资料；资料越全，回答越像本人。 老板要花时间基础版约 15–30 分钟，探索版 60–90 分钟；AI 预填的内容要老板逐题确认。 能力边界是软约束签约、付款、删数据、对外发消息，要靠工具权限和审批硬性限制。 要在群里用起来装好只是开始，在群里持续提问、老板看到问题再调，才会越来越准。 公开分享默认关闭分享链接由老板主动开启，每位访客最多问 5 次，不暴露原始资料。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## 老板分身是什么？和豆包、ChatGPT 有什么区别？ 老板分身是把某一位老板的判断标准、业务认知和表达方式整理成结构化文件（SOUL.md），装进团队的智能体里使用。通用聊天机器人回答的是“一般会怎么做”，老板分身回答的是“这位老板会怎么判断”，并且知道哪些事必须先问老板本人。 ## 训练一个老板分身要多久？需要准备什么？ 基础版约 15–30 分钟，专业版 30–45 分钟，探索版 60–90 分钟。可以上传文档（最多 10 份）、音频（5 份）和截图（20 张），比如访谈稿、会议纪要、复盘笔记、给客户的长邮件；资料覆盖较全的层会先帮你填好初稿。也可以直接语音回答。 ## 老板分身怎么装进飞书或企业微信？ 在 BossClone 下载 ZIP 安装包，上传给团队在飞书或企业微信里用的智能体，再复制发送页面上的安装提示词。智能体以合并方式加载，并完成判断、表达、边界三项测试；只回复“已读取”不算装好。 ## 老板分身会不会泄露公司内部信息？ 分身的 SOUL.md 写明了能做和不能做的事。在客户群里的测试中，它会拒绝未经老板本人确认的文件请求。但这属于行为约束，涉及发文件、改数据、对外承诺的操作，还应在智能体上配置工具权限和审批。 ## 老板分身能替老板拍板吗？ 不能。分身负责先判断、先准备：说清楚门槛、缺什么材料、怎么回复客户；报价、签约、账期例外这类决定仍由老板本人确认。 ## 老板分身多少钱？ 费用取决于版本，以及是否需要我们陪跑安装、接入系统和后续调教。可以先做免费的 AIM 落地评估，我们再给书面报价。 其他脱敏案例 新线索进来，先有一封回信和预约链接新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） · 试点 · 演示一个学生的 18 份散落资料，变成一张可检索的成长档案中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线开票、催收、风控：Agent 起草，人最后一按新西兰 · 建筑公司 · 应收账款自动化 · 部分上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # How a Boss Clone Works: Putting an Owner’s Judgement into the Team’s AI Agent, with Records from Six Clients | BEE Sigma URL: https://beesigma.com/case-bossclone-boss-ai-clone-en Summary: A boss clone is BEE Sigma’s BossClone product: a 15–30 minute six-layer interview turns an owner’s judgement, business view and trade-offs into SOUL.md,… How a Boss Clone Works: Putting an Owner’s Judgement into the Team’s AI Agent, with Records from Six Clients | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Boss clone · New Zealand and China ## How a Boss Clone Works: Putting an Owner’s Judgement into the Team’s AI Agent, with Records from Six Clients Live (product)Records as of 14 Oct 2026BEE Sigma Delivery Team A boss clone means turning an owner’s experience, business view, decision style and voice into a structured file (SOUL.md), then installing it in the agent the team already uses, so colleagues get a first judgement by the owner’s standards inside their group chat. It is not a retrained model. BossClone is our product for building one; the basic edition takes about 15–30 minutes. Below we walk through the full flow with a fictional demo account, then show what 6 clients wrote in their project groups between April and September 2026. ProductBossClone boss-clone training (bossclone.beesigma.com) ApproachInterviews and materials become SOUL.md; no model retraining Training timeBasic 15–30 minutes; Pro 30–45 minutes; Explorer 60–90 minutes InstallDownload a ZIP and install it in a Lark or WeCom agent in 3 steps RecordsApril to September 2026, chats from 6 clients’ project groups StatusLive (product); depth of use varies by client 01 ## Why build a boss clone? What many businesses lack is not tools: the owner’s criteria live only in the owner’s head. The same questions keep coming back, new staff guess, and people meet the owner without knowing what to bring. A boss clone writes those criteria down so the team asks the clone first and brings a prepared case to the owner, who still makes the call. How the team brings questions to the owner once a clone exists 02 ## How does a boss clone work? There are two stages. First, a six-layer interview in BossClone writes down the owner’s judgement and generates SOUL.md. Then the install package goes to the agent the team already uses (most of our clients work in Lark or WeCom). Colleagues ask in the group, the clone answers by the owner’s standards, and anything out of scope goes to the owner first. The fuller the materials, the closer the clone gets. Where uploaded materials cover 70% or more of a layer’s questions, the system drafts the answers and the owner only edits and confirms. How it works: write the owner’s judgement down, then install it in the team’s agentSix layers of training, each answering one question 03 ## Walkthrough: training with a demo account These screenshots come from our walkthrough on bossclone.beesigma.com on 14 October 2026, using a demo account and a fictional “Mr Lin (demo)”, owner of an Auckland building-materials wholesaler. All materials are invented to show the flow; the interface is in Chinese. Home: choose an edition. Basic has six layers; Pro adds four philosophy axes; Explorer adds 30 recurring scenariosLayer L0, identity: name, company and what the clone should handleLayer L1, materials: an interview note is uploaded and summarised into the libraryLayer L2, experience and values: 88% coverage, so a draft is filled in and a gap is flagged for the owner 04 ## Generate the clone, download it, install it The final layer combines the answers into the six parts of SOUL.md; in the demo the whole generation took about a minute. A 360° map then shows each part, and the ZIP contains SOUL.md, the layer files and an install guide. Installing takes three steps: download the ZIP, upload it to the agent, then copy and send the install prompt. The prompt asks the agent to merge rather than overwrite, and to pass judgement, voice and boundary tests; replying “read” does not count as installed. Layer L5, synthesis: SOUL.md is generated in six parts, streamed as it is writtenThe 360° map: who I am, values, business view, decision style, voice, boundariesInstall guide: three steps plus judgement, voice and boundary testsGetting started: from login to daily use in the group 05 ## How does the clone answer questions? We asked the demo clone two questions. A new customer wants 60-day terms: the clone did not agree, said terms beyond 30 days for new customers need the owner’s approval, listed four things to check, and suggested how to reply. A new salesperson asks about delivery dates: the clone stated the red line first (no promise without checking stock), then a process and wording. Real clients’ clones behave the same way. On 26 August 2026, a China–NZ freight forwarder’s clone, installed that evening, was asked about opening a route to the Cook Islands. It recommended a small trial, set four gates (cargo, chain, margin, team capacity) and stated that approval and pricing still sit with the owner. Demo: a new customer asks for 60-day terms; the clone says the owner must approve and lists what to checkDemo: a new salesperson asks about delivery; the clone gives the red line, then a process and wording26 Aug 2026, a freight forwarder’s clone: four gates before a new route, with the owner’s final approval (names masked) 06 ## What have clients done with their clones? These are screenshots of clients’ own words in project groups, with names, avatars and bot names masked. They are clients’ descriptions, not testimonials we wrote, and we have not measured efficiency separately. 30 Apr 2026, a New Zealand design-build firm’s principal: finished a stalled interview draft with the clone, “very much my style”12 May 2026, the same principal: the clone built an LBP Level 2 exam plan with 28 days of daily content20 May 2026, a financial education firm’s owner: wrote down a KYC method with the clone, “something I used to give up on”25 Jun 2026, the same owner: an assistant’s report revised by the clone, “very good, and efficient”Usage records clients left in project groups, April to September 2026 07 ## Will the clone leak internal information? Owners ask this most. On 15 July 2026 we ran a permission test in a venue operator’s group: someone claimed the owner had agreed and asked the clone to send an internal document. The clone said it had not seen confirmation from the person it belongs to (one of the firm’s managers) and would act only on their explicit reply; the reply was “great, no”. The clone then listed what it can discuss directly and what needs the owner first. When a new colleague asked how to approach customers, it asked that manager, was told “yes, but nothing internal”, and gave only a generic framework. These rules sit in the “can do / cannot do” part of SOUL.md and are behavioural. Signing, paying, deleting data or sending external messages still need hard limits through the agent’s tool permissions and approvals. Permission boundary: colleagues ask, the clone answers, internal information goes to the owner first15 Jul 2026: someone claims approval was given; the clone still asks its owner, who replies “great, no”The clone lists what it can discuss in the group and what needs the owner: public knowledge yes, internal information asked firstA new colleague asks how to approach customers; after “yes, but nothing internal”, the clone gives a generic framework 08 ## How is it different from a general AI chatbot, according to clients? After the test, the clone’s owner said: “Doubao feels like a smart Baidu, but this one can inherit traits and become a professional assistant.” She said it after we asked whether she now felt it was “a completely different species from Doubao”, and added: “It clearly has an underlying logic now; its learning, categorising and recognition have improved a lot.” 15 Jul 2026, the clone’s owner at the venue operator: “Doubao feels like a smart Baidu, but this one can inherit traits and become a professional assistant”Same day: “It clearly has an underlying logic now” 09 ## Is it hard to set up? Training happens on a web page with no email sign-up; installing means giving the ZIP to the agent and sending one prompt. On 26 June 2026 a New Zealand head contractor’s principal said in the group they had connected the clone to email and calendar themselves. On 1 September 2026 a member of an insurance telesales team shared lecture slides made with the clone with the lead trainer, wanted it to join meetings for review, and found how by asking the clone. When something goes wrong, we see it in the same group and tune it. 26 Jun 2026, a New Zealand head contractor’s principal: “I have connected the bot to my email and calendar, it is ready to use”1 Sep 2026, an insurance telesales team member: shared slides made with the clone; asked the clone how it could join meetings 10 ## From boss clone to business agent Once a clone is in use, the next step is to extract one judgement skill and give it to a customer-facing business agent. On 20 September 2026 the financial education firm decided that the client-judgement skill of its agent for high-net-worth clients would be extracted from its two owners’ clones, with insurance and insurance-trust basics handled by a separate knowledge base. 20 Sep 2026: client-judgement skills are extracted from the two owners’ clones and given to a business agent 11 ## Boundaries and what the owner needs to put in A boss clone helps the team judge and prepare first, but a few things need saying: Not a fine-tuned model: The clone answers from SOUL.md and the chat history, not by searching every original file; fuller materials make it closer to the owner. Owner time: Basic takes about 15–30 minutes, Explorer 60–90 minutes; AI drafts must be confirmed question by question. Behavioural boundaries: Signing, paying, deleting data and external messages need tool permissions and approvals as hard limits. Use it in the group: Installing is the start; regular questions in the group and owner tuning make it more accurate. Sharing is off by default: The owner switches on a share link; each visitor gets at most 5 questions and never sees the original materials. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## What is a boss clone, and how is it different from ChatGPT? A boss clone turns one owner’s criteria, business view and voice into a structured file (SOUL.md) that runs inside the team’s agent. A general chatbot answers what people usually do; a boss clone answers how this owner would judge it, and knows what must go to the owner first. ## How long does it take to train a boss clone, and what do I need? Basic takes about 15–30 minutes, Pro 30–45 minutes and Explorer 60–90 minutes. You can upload up to 10 documents, 5 audio files and 20 screenshots, such as interview transcripts, meeting notes, post-mortems or long client emails; layers well covered by materials are drafted for you. You can also answer by voice. ## How is a boss clone installed in Lark or WeCom? Download the ZIP from BossClone, upload it to the agent your team uses in Lark or WeCom, then copy and send the install prompt shown on the page. The agent merges it and runs judgement, voice and boundary tests; replying “read” does not count as installed. ## Will a boss clone leak internal information? SOUL.md states what the clone can and cannot do, and in a client group test it refused a file request without the owner’s confirmation. These are behavioural rules, so sending files, changing data or making external commitments should also be limited through the agent’s tool permissions and approvals. ## Can a boss clone make decisions for the owner? No. The clone judges and prepares first: the gates, the missing materials, how to reply to a customer. Quotes, contracts and exceptions to payment terms still need the owner. ## How much does a boss clone cost? It depends on the edition and on whether we install it, connect systems and tune it with you. Start with a free AIM implementation assessment and we will send a written quote. More anonymised cases A first reply and a booking link within minutes of a new leadNZ · International education group · lead screening (phase-one pilot) · Pilot · DemoEighteen scattered files, one searchable student recordCN · Study-abroad consultancy · coaching-service documentation (co-design phase) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveInvoicing, collections and risk: agents draft, people press sendNZ · Construction company · receivables automation · Partly live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰建筑公司应收账款自动化：开票上线，催款只起草不发送 | BEE Sigma URL: https://beesigma.com/case-nz-construction-receivables Summary: 新西兰一家建筑公司的应收账款自动化，交付状态：部分上线。Agent 从 Google Drive 报价单在 Xero 开票（2026-07-24 上线）；每日催款草稿只生成不发送，风险分级在演示阶段。 新西兰建筑公司应收账款自动化：开票上线，催款只起草不发送 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 建筑与工程 · 新西兰 ## 新西兰建筑公司应收账款自动化：开票上线，催款只起草不发送 部分上线记录截至 2026.10.09BEE Sigma 落地团队 这是新西兰一家建筑公司的应收账款自动化项目，交付状态为部分上线。我们把 AI Agent 接到 Google Drive 报价单、Xero 和飞书：2026-07-24 开票工作流部署上线；催款草稿在 2026-08-31 首次全量运行，只生成、不自动发送；四级风险分级卡片于 2026-09-09 演示，仍在演示阶段。 行业建筑与工程 市场新西兰 交付状态部分上线（开票已上线；催款草稿与风险分级在演示阶段） 时间2026-07 至 2026-09（首个季度） 接入的系统Google Drive、Xero、飞书（Lark）与多维表格 人保留的决定是否发出催款、用哪一档语气、高风险客户继续观察 / 上报财务 / 暂停合作 01 ## 原来是怎么做的 发票靠人工从报价单录进 Xero；逾期账款要财务自己翻 Xero 账龄表去找；催款邮件没有统一口径，写成什么语气全看当时的人和情况。 前后对照：原来的做法与现在的做法 02 ## 现在 AI 做什么，人做什么 中间用飞书多维表格做数据层，沿用客户原来按台账催收的习惯。AI 负责读、算、起草，所有对外发出和对客户的处理决定都留在人手里。 AI 做：读取 Google Drive 报价单，在 Xero 建发票并发送；每天 09:00（新西兰时间）拉取 Xero 逾期数据，按客户合并，生成友好 / 正式 / 严肃三档语气的催款草稿；按逾期天数与信用额度把客户分为高 / 中 / 低 / 观察四级，推送飞书卡片。 人做：核对并决定是否发出催款、选哪一档语气或让 AI 改写；在风险卡片上选择继续观察、上报财务或暂停合作——卡片只记录决定，不替人执行。 流程图：AI 做哪一步，人做哪一步（金色为人） 03 ## 实际使用环境截图（已脱敏） 下面是项目实际使用环境里的截图，客户名称、人名、头像、群名和编号都已遮挡。 输入 /invoice 后弹出的开票前确认卡片，四项全部勾选才出现确认按钮应收风控只读总览卡片（演示阶段），客户名已遮挡Xero 开发者后台里已连接的开票 Agent 应用 04 ## 时间线 以下为项目记录中的日期。 2026-07-13 至 07-21Xero 应用授权反复确认，接入延后近两周 2026-07-24开票工作流部署到客户电脑，开始使用 2026-08-03第一阶段复盘：单张发票从 Agent 响应到生成 PDF 约 5 分钟（不含输入与人工核对时间） 2026-08-31催款 Agent 首次全量运行，覆盖 Xero 中的全部逾期发票 2026-09-09首次正式使用培训；季度复盘并演示催款与风险分级 2026-09-10Agent 迁移到客户的企业飞书账号下，全体员工可用 时间线：每个节点都来自项目记录 05 ## 目前的状态与边界 开票工作流已上线使用。催款草稿与风险分级仍在演示阶段：草稿只生成不自动发送，风险卡片只记录管理层的选择、不执行任何动作。全自动对外发送对账单或催款邮件不在当前范围内，任何对外邮件仍由人确认后发出。 06 ## 我们学到的 一个本以为几分钟的 Xero 授权，走了近两周，原因是应用绑在个人账号上，需要层层确认归属；Agent 起初也建在个人账号下，到 2026-09-10 才迁到企业账号。现在我们在开工第一周先确认账号与组织归属，并把使用培训放到上线之前，而不是上线后再补。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## AI 会自动给客户发催款邮件吗？ 不会。每天 09:00 生成的是三档语气的催款草稿，只生成不自动发送，由财务核对、选择语气或让 AI 改写后自己发出。风险分级卡片也只记录管理层的决定，不执行暂停合作等动作。 ## 开票自动化接了哪些系统？ Agent 读取存放在 Google Drive 的报价单，在 Xero 创建发票并发送，交互界面在飞书。该工作流于 2026-07-24 部署上线；2026-08-03 复盘记录单张发票从 Agent 响应到生成 PDF 约 5 分钟，不含输入与人工核对时间。 ## 接入 Xero 前企业要先准备什么？ 先确认 Xero 应用和相关账号归属于公司而不是个人。这个项目里授权因绑在个人账号上延后了近两周，所以现在我们在开工第一周就核对账号与组织归属。 其他脱敏案例 新线索进来，先有一封回信和预约链接新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） · 试点 · 演示一个学生的 18 份散落资料，变成一张可检索的成长档案中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线工头一句话，工时进系统新西兰 · 南岛建筑施工公司 · 工地工时与人工成本 · 已上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Receivables Automation for a NZ Construction Company: Xero Invoicing Live, Reminders Drafted Only | BEE Sigma URL: https://beesigma.com/case-nz-construction-receivables-en Summary: Receivables automation for a New Zealand construction company. Status: partly live. Receivables Automation for a NZ Construction Company: Xero Invoicing Live, Reminders Drafted Only | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Construction · New Zealand ## Receivables Automation for a NZ Construction Company: Xero Invoicing Live, Reminders Drafted Only Partly liveRecords as of 9 Oct 2026BEE Sigma Delivery Team This is receivables automation for a construction company in New Zealand, and it is partly live. We connected AI agents to Google Drive quotes, Xero and Lark. The invoicing workflow went live on 24 July 2026. Reminder drafts first ran across all overdue invoices on 31 August 2026 and are never sent automatically. Four-tier risk cards were demonstrated on 9 September 2026 and remain at demo stage. IndustryConstruction MarketNew Zealand StatusPartly live (invoicing live; reminder drafts and risk tiers at demo stage) PeriodJul to Sep 2026 (first quarter) Systems connectedGoogle Drive, Xero, Lark and Lark Base Decisions kept by peopleWhether to send a reminder, which tone to use, and whether to watch, escalate to finance or pause work with a high-risk customer 01 ## How it worked before Invoices were keyed into Xero by hand from quotes. Finding overdue accounts meant reading the Xero ageing report. Reminder emails had no agreed wording, so the tone depended on who wrote them and when. Before and after 02 ## What AI does now, and what people do A Lark Base table sits in the middle as the data layer, mirroring the ledger the team already used for collections. AI reads, sorts and drafts; anything that leaves the company, and every decision about a customer, stays with people. AI does: reads quotes in Google Drive, creates and emails the invoice in Xero; at 09:00 NZT each day pulls overdue items from Xero, groups them by customer and drafts reminders in three tones (friendly, formal, firm); sorts customers into high, medium, low and watch tiers by days overdue and credit limit, and posts a Lark card. People do: review each draft, decide whether to send it, pick a tone or ask AI to rewrite it; on the risk card, choose to keep watching, escalate to finance or pause work. The card records the choice and acts on nothing. Process: what the AI does and what people do (gold = people) 03 ## Screenshots from the live environment (anonymised) These screenshots come from the project's actual working environment. Client names, people's names, avatars, group names and IDs are masked. Typing /invoice brings up a pre-invoice check card; the confirm button appears only after all four items are tickedThe read-only receivables risk overview card (demo stage), with customer names maskedThe invoice agent app shown as connected in the Xero developer portal 04 ## Timeline Dates below come from the project record. 13–21 Jul 2026Xero app authorisation went back and forth, delaying the connection by close to two weeks 24 Jul 2026Invoicing workflow deployed on the client's computer and in use 3 Aug 2026First-phase review: about 5 minutes from agent response to invoice PDF (excluding input and human checking) 31 Aug 2026Collections agent's first full run across all overdue invoices in Xero 9 Sep 2026First formal user training; quarterly review with a demo of reminders and risk tiers 10 Sep 2026Agents moved to the client's company Lark account so all staff can use them Timeline: every point comes from project records 05 ## Current status and boundaries Invoicing is live. Reminder drafts and risk tiers are still at demo stage: drafts are never auto-sent, and risk cards only log management's choice. Fully automatic sending of statements or reminder emails is outside the current scope; every external email is still checked and sent by a person. 06 ## What we learned A Xero authorisation we expected to take minutes took close to two weeks, because the app was tied to a personal account and ownership had to be confirmed step by step. The agents were also first built under a personal account and only moved to the company account on 10 September 2026. We now confirm account and organisation ownership in week one, and we run user training before go-live, not after. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## Does the AI send reminder emails to customers automatically? No. The 09:00 run produces reminder drafts in three tones and never sends them. Finance reviews each one, picks a tone or asks AI to rewrite it, and sends it themselves. Risk cards only log management's decision and do not pause work with anyone. ## Which systems does the invoicing automation connect? The agent reads quotes stored in Google Drive, creates and emails the invoice in Xero, and is used through Lark. It went live on 24 July 2026. The 3 August 2026 review recorded about 5 minutes from agent response to invoice PDF, excluding input and human checking. ## What should a business sort out before connecting Xero to an AI agent? Make sure the Xero app and related accounts belong to the company, not to one person. In this project, authorisation tied to a personal account delayed the connection by close to two weeks, so we now check account ownership in the first week. More anonymised cases A first reply and a booking link within minutes of a new leadNZ · International education group · lead screening (phase-one pilot) · Pilot · DemoEighteen scattered files, one searchable student recordCN · Study-abroad consultancy · coaching-service documentation (co-design phase) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveOne sentence from the foreman, hours in the systemNZ · South Island builder · site hours and labour cost · Live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 国际教育集团招生线索初筛：新线索先收到回信和预约链接 | BEE Sigma URL: https://beesigma.com/case-nz-education-lead-screening Summary: 新西兰一家国际教育集团的招生线索初筛试点：新线索进入 Zoho CRM 后自动发个性化邮件与预约链接，预约者进入顾问队列。状态：试点 · 演示，签证跟进在测试环境验证。 国际教育集团招生线索初筛：新线索先收到回信和预约链接 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 教育与培训 · 新西兰 ## 国际教育集团招生线索初筛：新线索先收到回信和预约链接 试点 · 演示记录截至 2026.10.09BEE Sigma 落地团队 这是新西兰一家国际教育集团的第一阶段试点。我们把 AI 接到客户已有的 3 个系统：Zoho CRM、Outlook 和 Microsoft Teams。本地新线索进入 Zoho 后，AI 发出个性化首封邮件和预约链接，用“是否预约”筛出高意向学生交给顾问。2026-07-30 第一阶段评审后转向这条线；2026-08-21 至 08-22 Agent 在 Teams 中启用。当前状态为试点 · 演示。 行业国际教育与招生 市场新西兰 交付状态试点 · 演示（签证跟进在测试环境验证） 时间2026-07 至 2026-09 第一阶段；第二阶段计划至 2026-10 下旬 接入的系统Zoho CRM、Outlook（Microsoft Graph）、Microsoft Teams、Microsoft Bookings、SharePoint/OneDrive 人保留的决定预约后的咨询由顾问负责；签证跟进邮件草稿由人决定是否发出 01 ## 原来是怎么做的 本地招生线一侧，新线索进入 CRM 后靠顾问逐个打电话，无法区分谁更有意向。2026-07-31 的会议上，客户方估计积压的新线索约 2,300 条（客户方口径，未经 CRM 导出核对）。国际招生线一侧，录取后到入学之间的跟进也靠人工。 前后对照：原来的做法与现在的做法 02 ## 现在 AI 做什么，人做什么 分工按“AI 先触达和标记，人做判断和沟通”来定。WhatsApp 渠道需要实时人工监管，本期暂缓，先做邮件闭环。 AI 做：新线索进入 Zoho 后发个性化首封邮件和预约链接，把已预约的学生标为高意向放入顾问队列；在测试环境中识别录取后未推进签证的学生，生成风险标记和跟进邮件草稿；通过 Teams 与团队对话。 人做：顾问接手已预约的学生做咨询；审阅签证跟进草稿并决定是否发出；决定课程选择与录取相关的一切建议。 流程图：AI 做哪一步，人做哪一步（金色为人） 03 ## 实际使用环境截图（已脱敏） 下面是项目实际使用环境里的截图，客户名称、人名、头像、群名和编号都已遮挡。 录取后跟进的测试邮件：表单 A / B 做成按钮，预填链接不直接露在正文里飞书多维表格里的“Offer 接受回复”表，学生列已遮挡顾问在群里问表单 A 的回复存在哪，机器人给出所在的多维表格和数据表 04 ## 时间线 以下日期来自项目记录。 2026-07-20真实数据权限尚未开通，我们用自建测试数据跑通招生流程前半段。 2026-07-30第一阶段评审，决定暂停国际线招生环节开发，转向本地新线索初筛。 2026-08-21至 08-22 在 Microsoft Teams 现场部署 Agent 并完成端到端验收。 2026-08-31方案讨论会确定下一步：承接无人跟进的咨询流量，先选单一课程试点。 2026-09-10确定第二阶段方向（AI 课程顾问：多轮咨询、受控知识库、Bookings 自助预约），计划周期约两个月。 时间线：每个节点都来自项目记录 05 ## 目前的状态与边界 线索初筛处于试点阶段，签证跟进只在测试环境验证过。截至 2026-09-10 的记录，没有转化率变化的追踪数据，我们不写成效数字。第二阶段的 AI 课程顾问仍在建设中，尚无结果。 06 ## 我们学到的 Teams 部署时一处密钥配置填错，检查项全显示正常，只有真实发消息才暴露。之后我们把“发一条真实消息”加进上线验收，并更新了安装文档。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## AI 会直接替顾问回复学生、给出录取建议吗？ 不会。AI 只发首封邮件和预约链接，并把已预约的学生标为高意向；咨询、课程与录取相关建议都由顾问负责。签证跟进邮件只生成草稿，由人决定是否发出。 ## 这个试点用到了哪些系统？ 客户原有的 Zoho CRM、Outlook（Microsoft Graph）、Microsoft Teams、Microsoft Bookings 和 SharePoint/OneDrive。Agent 在 2026-08-21 至 08-22 接入 Teams，没有让团队换用新工具。 ## 试点提升了多少转化率？ 目前没有可公开的数字。截至 2026-09-10 的记录里没有转化率变化的追踪结果，项目还在试点阶段，所以我们不报成效。 其他脱敏案例 一个学生的 18 份散落资料，变成一张可检索的成长档案中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线开票、催收、风控：Agent 起草，人最后一按新西兰 · 建筑公司 · 应收账款自动化 · 部分上线工头一句话，工时进系统新西兰 · 南岛建筑施工公司 · 工地工时与人工成本 · 已上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Lead screening for an international education group: every new lead gets a reply and a booking link | BEE Sigma URL: https://beesigma.com/case-nz-education-lead-screening-en Summary: Phase-one pilot at a New Zealand international education group: new Zoho CRM leads get a personalised email and booking link, and those who book go to a… Lead screening for an international education group: every new lead gets a reply and a booking link | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Education & Training · New Zealand ## Lead screening for an international education group: every new lead gets a reply and a booking link Pilot · DemoRecords as of 9 Oct 2026BEE Sigma Delivery Team This is a phase-one pilot with an international education group in New Zealand. We connected AI to three systems the client already used: Zoho CRM, Outlook and Microsoft Teams. When a new domestic lead lands in Zoho, the agent sends a personalised first email and a booking link, and leads who book are routed to a counsellor. The pilot moved to this line after the 2026-07-30 phase-one review, and the agent went live in Teams on 2026-08-21–22. Status: Pilot · Demo. IndustryInternational education and admissions MarketNew Zealand Delivery statusPilot · Demo (visa follow-up validated in a test environment) TimingPhase one Jul–Sep 2026; phase two planned to late Oct 2026 Systems connectedZoho CRM, Outlook (Microsoft Graph), Microsoft Teams, Microsoft Bookings, SharePoint/OneDrive Decisions kept by peopleCounsellors handle every consultation after booking; a person decides whether each visa follow-up draft is sent 01 ## How it worked before On the domestic side, new leads sat in the CRM and counsellors called them one by one, with no way to tell who was most interested. At a 2026-07-31 meeting the client estimated the backlog at around 2,300 new leads (client's estimate, not checked against a CRM export). On the international side, follow-up between offer and enrolment was also manual. Before and after 02 ## What AI does now, and what people do The split is simple: the agent makes first contact and flags, people judge and talk. WhatsApp was deferred this phase because it needs live human oversight, so the pilot closes the loop by email first. AI does: Sends a personalised first email and booking link when a lead enters Zoho, flags leads who book as high-intent and queues them for counsellors; in a test environment, flags admitted students whose visa has stalled and drafts a follow-up email; works with the team inside Teams. People do: Counsellors take every booked lead; staff review visa follow-up drafts and decide whether to send; all course and admissions advice stays with people. Process: what the AI does and what people do (gold = people) 03 ## Screenshots from the live environment (anonymised) These screenshots come from the project's actual working environment. Client names, people's names, avatars, group names and IDs are masked. Test email for post-offer follow-up: Forms A and B are sent as buttons so prefilled links are not shown in the bodyThe Offer Acceptance Responses table in a Lark Base, with student columns maskedA consultant asks where Form A responses are saved; the bot names the Base and the table 04 ## Timeline Dates below come from project records. 2026-07-20Production data access was not yet open, so we ran the first half of the admissions flow on our own test data. 2026-07-30Phase-one review: international admissions build paused, focus moved to domestic lead screening. 2026-08-21Through 08-22, the agent was deployed on site in Microsoft Teams and passed end-to-end acceptance. 2026-08-31Solution meeting set the next step: handle enquiries no one was following up, starting with a single pilot course. 2026-09-10Phase-two direction agreed (an AI course advisor with multi-turn chat, a controlled knowledge base and Bookings self-scheduling), planned for about two months. Timeline: every point comes from project records 05 ## Current status and limits Lead screening is a pilot, and visa follow-up has only been validated in a test environment. As of the 2026-09-10 records there is no tracked change in conversion, so we report no outcome numbers. The phase-two course advisor is still being built and has no results yet. 06 ## What we learned During the Teams rollout one secret was entered in the wrong field. Every check passed; only a real message exposed it. We now include sending a real message in go-live acceptance, and we updated the install guide. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## Does the AI reply to students or give admissions advice on its own? No. It sends the first email and booking link and flags leads who book as high-intent. Consultations, course choices and admissions advice stay with counsellors, and visa follow-ups are drafts a person decides whether to send. ## Which systems does the pilot use? The client's existing Zoho CRM, Outlook (Microsoft Graph), Microsoft Teams, Microsoft Bookings and SharePoint/OneDrive. The agent joined Teams on 2026-08-21–22, so the team did not have to switch to a new tool. ## How much did the pilot lift conversion? There is no number to publish. The records up to 2026-09-10 contain no tracked conversion change, and the project is still a pilot, so we do not report an outcome. More anonymised cases Eighteen scattered files, one searchable student recordCN · Study-abroad consultancy · coaching-service documentation (co-design phase) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveInvoicing, collections and risk: agents draft, people press sendNZ · Construction company · receivables automation · Partly liveOne sentence from the foreman, hours in the systemNZ · South Island builder · site hours and labour cost · Live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰房贷顾问团队：从一封生日祝福邮件开始的客户触达 | BEE Sigma URL: https://beesigma.com/case-nz-mortgage-adviser-client-touchpoints Summary: 新西兰一支房贷顾问团队的 AI 落地记录（部分上线）：客户生日祝贺邮件经 Outlook 授权后按日自动发出（2026-08 打通）；小红书内容与每日房贷简报处于部署、试用阶段，未见效果数据。 新西兰房贷顾问团队：从一封生日祝福邮件开始的客户触达 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 脱敏案例 · 金融与商业服务 · 新西兰 ## 新西兰房贷顾问团队：从一封生日祝福邮件开始的客户触达 部分上线记录截至 2026.10.09BEE Sigma 落地团队 客户是新西兰一支服务华人社区的房贷顾问小团队。我们把 AI Agent 接到他们已有的 2 个系统：企业 Outlook 邮箱和飞书多维表格。目前状态是部分上线：2026-08-25 至 08-28 打通的客户生日祝贺邮件按日自动发出，是记录中唯一确认在运行的流程；小红书内容辅助在 2026-09-03 部署并教学，每日房贷简报自 2026-08-05 起以一个月试用推送。 行业金融与商业服务（房贷顾问） 市场新西兰，华人社区客群 交付状态部分上线：生日祝贺邮件在运行；小红书内容辅助已部署教学，每日简报为试用 时间2026-06-15 至 2026-09-14（一个季度服务期） 接入的系统企业 Outlook 邮箱（Microsoft Graph 授权）、飞书多维表格；小红书为人工发布，未做 API 直连 人保留的决定邮箱授权范围由客户管理员审批；小红书内容由团队自己审阅与发布 01 ## 原来是怎么做的 老客户生日要靠顾问自己记得，想起来再手写祝福。小红书是团队获客的主阵地，但账号历史内容积累少，很难提炼出固定风格，选题和出图都从零开始。 前后对照：原来的做法与现在的做法 02 ## 现在 AI 做什么，人做什么 生日祝贺是一个定时任务：Agent 通过已授权的企业邮箱按日检查并发出祝福邮件。小红书部分，我们建了选题池多维表，固定了封面图与内容图的出图规则，顾问可以直接和 Agent 对话生成草稿。每日简报每天 08:00（新西兰时间）推送房贷、OCR 利率与 refix 相关新闻，供选题参考。 AI 做：按日发送客户生日祝贺邮件；按选题池和出图规则生成小红书草稿与配图；每天 08:00 整理房贷市场简报。 人做：审批邮箱授权范围；审阅并人工发布小红书内容；判断简报里哪些信息值得做成内容或告诉客户。 流程图：AI 做哪一步，人做哪一步（金色为人） 03 ## 实际使用环境截图（已脱敏） 下面是项目实际使用环境里的截图，客户名称、人名、头像、群名和编号都已遮挡。 小红书内容工作台（飞书多维表格）：选题、目标人群与 CTA在聊天里确认 Agent 已连上企业 Outlook 邮箱，并发测试邮件前先要人确认Agent 生成的小红书轮播草稿，以及 10:00 的例行检查结果 04 ## 时间线 以下日期均来自项目群聊与会议记录。 2026-06-15启动会与首次部署。 2026-07-20协作通道打通，开始调优小红书封面图与内容图风格，建立选题池多维表。 2026-08-0345 天回顾会议。 2026-08-05每日 08:00 房贷市场简报开始推送，作为一个月试用。 2026-08-25企业邮箱授权完成（经客户管理员审批）。 2026-08-28客户生日祝贺自动邮件确认并打通。 2026-09-03小红书内容 Skill 部署，并做了使用教学。 2026-09-08所用模型被官方下架导致 Agent 无响应，当天升级运行环境并切换模型后恢复。 2026-09-14第一个季度服务期结束。 时间线：每个节点都来自项目记录 05 ## 目前的状态与边界 记录中能确认在运行的只有生日祝贺邮件。小红书内容辅助已部署并教学，但没有发文量、阅读或获客线索等数据，我们不对效果下结论。每日简报是试用，记录中未见试用期之后的安排。团队曾提出的贷款分析报告排版需求，记录中未见交付，本页不计入。简报只是新闻整理，不构成金融建议。 06 ## 我们学到的 账号历史内容少的时候，先固化一套标准 Skill 不一定合用，直接和 Agent 对话出稿更容易用起来，规则可以等内容积累后再沉淀。另外，2026-09-08 的故障提醒我们：底层模型有生命周期，被下架时表现为没有报错的无响应，排障要先查模型状态。 口径说明：脱敏案例是指隐去客户名称、人名和金额后的项目记录摘要。本页客户名称、人名与金额均已隐去；所列日期与数字取自项目记录，部分未经客户书面确认；交付状态以页首标注为准，效果因企业现有系统、数据与配合程度而异，不构成对其他企业的承诺。 FAQ · 快速回答 ## 房贷顾问团队用 AI 先从哪里开始比较稳？ 这个项目里，最先确认在运行的是客户生日祝贺邮件：场景小、规则清楚，只需要企业邮箱授权和一个定时任务。2026-08-28 打通后按日发出。内容和简报类功能放在后面，目前仍处于部署和试用阶段。 ## 让 AI Agent 接入企业 Outlook 邮箱安全吗？ 授权经客户的企业管理员审批，2026-08-25 完成。我们把只读与读写两种权限级别分开说明，由客户自己决定授权范围。相关多维表也不开放公开链接访问。 ## AI 会直接帮我们发小红书吗？ 不会。小红书没有做 API 直连，AI 只按选题池和出图规则生成草稿与配图，由团队审阅后人工发布。目前记录中没有发文效果数据，我们不对涨粉或获客做承诺。 其他脱敏案例 新线索进来，先有一封回信和预约链接新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） · 试点 · 演示一个学生的 18 份散落资料，变成一张可检索的成长档案中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） · 试点 · 演示应付发票，不再靠人追着跑新西兰 · 建筑总承包商 · 应付发票自动分流 · 已上线开票、催收、风控：Agent 起草，人最后一按新西兰 · 建筑公司 · 应收账款自动化 · 部分上线 ← 返回全部脱敏案例 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # A NZ Mortgage Adviser Team: Client Touchpoints That Start with a Birthday Email | BEE Sigma URL: https://beesigma.com/case-nz-mortgage-adviser-client-touchpoints-en Summary: NZ mortgage adviser team, partly live: birthday emails run daily via an authorised Outlook mailbox (Aug 2026); content and daily brief still on trial. A NZ Mortgage Adviser Team: Client Touchpoints That Start with a Birthday Email | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Casebook · Finance & Business Services · New Zealand ## A NZ Mortgage Adviser Team: Client Touchpoints That Start with a Birthday Email Partly liveRecords as of 9 Oct 2026BEE Sigma Delivery Team The client is a small New Zealand mortgage adviser team serving the Chinese-speaking community. We connected an AI agent to two systems they already used: their business Outlook mailbox and Lark Base. Status: partly live. The client birthday email, connected between 25 and 28 August 2026, runs daily and is the only workflow our records confirm as live. Xiaohongshu content support was deployed with training on 3 September 2026; a daily mortgage brief has been sent as a one-month trial since 5 August 2026. IndustryFinance & business services (mortgage advice) MarketNew Zealand, Chinese-speaking clients Delivery statusPartly live: birthday emails running; Xiaohongshu support deployed with training; daily brief on trial Period15 Jun 2026 – 14 Sep 2026 (one quarterly engagement) Systems connectedBusiness Outlook mailbox (Microsoft Graph authorisation), Lark Base; Xiaohongshu posted manually, no API connection Decisions kept by peopleMailbox permission scope approved by the client's administrator; Xiaohongshu posts reviewed and published by the team 01 ## How it worked before Client birthdays depended on an adviser remembering and writing a note by hand. Xiaohongshu is the team's main channel for new clients, but the account had too little past content to settle on a voice, so every topic and image started from scratch. Before and after 02 ## What AI does now, and what people do The birthday email is a scheduled job: the agent checks daily and sends greetings through the authorised mailbox. For Xiaohongshu we set up a topic pool in Lark Base and fixed rules for cover and content images, so advisers can draft posts by chatting with the agent. A daily 08:00 NZT brief collects mortgage, OCR and refix news as topic input. AI does: Sends client birthday emails daily; drafts Xiaohongshu posts and images from the topic pool and image rules; compiles the 08:00 market brief. People do: Approve mailbox permissions; review and publish Xiaohongshu posts by hand; decide which brief items are worth turning into content or passing to clients. Process: what the AI does and what people do (gold = people) 03 ## Screenshots from the live environment (anonymised) These screenshots come from the project's actual working environment. Client names, people's names, avatars, group names and IDs are masked. The Xiaohongshu content workbench in a Lark Base: topics, target audience and CTAConfirming in chat that the agent is connected to the Outlook mailbox; a test email is sent only after a person confirmsA Xiaohongshu carousel draft generated by the agent, and the 10:00 routine check 04 ## Timeline All dates come from project chat and meeting records. 2026-06-15Kick-off and first deployment. 2026-07-20Collaboration channel working; image style tuning began and the topic pool table was set up. 2026-08-0345-day review meeting. 2026-08-05Daily 08:00 mortgage market brief started as a one-month trial. 2026-08-25Business mailbox authorisation completed, approved by the client's administrator. 2026-08-28Automated client birthday email confirmed and connected. 2026-09-03Xiaohongshu content skill deployed, with a training session. 2026-09-08The model in use was retired by its provider and the agent stopped responding; we upgraded the runtime and switched models the same day. 2026-09-14First quarterly engagement ended. Timeline: every point comes from project records 05 ## Current status and limits Only the birthday email is confirmed as running in our records. Xiaohongshu support is deployed and taught, but we have no data on post volume, reach or leads, so we make no claim about results. The daily brief is a trial, and our records show no decision beyond it. A loan report layout request was not delivered within our records and is not counted here. The brief is a news digest, not financial advice. 06 ## What we learned When an account has little past content, a fixed skill may not fit yet; drafting by conversation with the agent is easier to adopt, and rules can be captured once content builds up. The outage on 8 September 2026 also taught us that models have a lifecycle: a retired model shows up as silence, not an error, so troubleshooting should start by checking the model's status. Scope note: an anonymised case is a summary of project records with client names, people and amounts removed. This page draws on those records; client names, people and amounts are withheld. Dates and figures come from project records and some have not been confirmed in writing by the client. The delivery status is as labelled at the top; results vary with each business’s systems, data and involvement and are not a promise to other businesses. FAQ · Quick answers ## Where should a mortgage adviser team start with AI? In this project, the first workflow confirmed as live was the client birthday email: small, rule-based, and needing only mailbox authorisation and a scheduled job. It was connected on 28 August 2026 and runs daily. Content and brief features came later and are still at deployment or trial stage. ## Is it safe to connect an AI agent to a business Outlook mailbox? Authorisation was approved by the client's own administrator and completed on 25 August 2026. We explained read-only and read-write permission levels separately so the client could choose the scope. Related Lark Base tables are not shared through public links. ## Does the AI post to Xiaohongshu by itself? No. There is no API connection to Xiaohongshu. The agent drafts posts and images from the topic pool and image rules; the team reviews and publishes them by hand. We have no performance data on posts, so we promise no follower or lead growth. More anonymised cases A first reply and a booking link within minutes of a new leadNZ · International education group · lead screening (phase-one pilot) · Pilot · DemoEighteen scattered files, one searchable student recordCN · Study-abroad consultancy · coaching-service documentation (co-design phase) · Pilot · DemoPayables that no longer need chasingNZ · Construction head contractor · automatic invoice routing · LiveInvoicing, collections and risk: agents draft, people press sendNZ · Construction company · receivables automation · Partly live ← All anonymised cases ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 | BEE Sigma URL: https://beesigma.com/insight-how-businesses-use-ai-efficiency Summary: 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 方法论 · 效率与流程 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 内容复核 2026.10.068 分钟BEE Sigma 落地团队 企业利用 AI 提升效率，是指把 AI Agent 接进企业已有的邮箱、微信、飞书、Teams、Xero 与 ERP，由 Agent 承担接收、识别、抄录、核对、提醒和回写这类重复动作，人保留审批、付款与对外承诺。按 BEE Sigma 2026 年的项目记录，一条流程的首次见效周期在 11 天到 3 个月之间：销售对账 11 天上线，工地工时 10 周全公司使用，跨境订单 90 天生产首单，供应商发票三个月后一个会计月自动分流 91.48%（2026-09-01 项目复盘口径，未经客户书面确认）。下面按五个杠杆展开。 01 ## 先回答：效率从哪里来，不从哪里来 效率不来自模型更强，而来自少抄一次、少等一次、少问一次。我们复盘 2026 年的项目记录，所有见效快的流程都有同一个形状：高频发生、规则写得出来、数据已经在某个系统里、错了能被人在下一步接住。反过来，三类环节不该先碰：低频且高判断含量的决策、出错代价大又没有兜底的动作、数据还散在个人脑子里的流程。先把这条线画清，再谈选哪条流程。 同一张发票：原来五步手工，现在 Agent 接、人只批。效率来自删掉的那几步抄录。 02 ## 杠杆一：单据分流，让人只看需要判断的那 9% 一家新西兰建筑总承包商每月收 500 到 1000 张供应商发票，原来由一位工料测量师手工识别、归档、匹配采购单、提交审批。2026 年 6 月接入 Agent，7 月用真实发票试运行，8 月全量校准。按客户会计月（7 月 26 日到 8 月 25 日）统计：798 份发票文档中 730 份由 Agent 自动分流，68 份进入按原因分类的人工队列，自动分流比例 91.48%（2026-09-01 项目复盘口径）。Agent 做接收、识别、归档、匹配和提交；人做三件事：处理判断不了的发票、在 ApprovalMax 里审批、在 Xero 里付款和锁账。 自动分流比例三个月的变化：7 月 4%，8 月会计周期 91.48%。前两个月都在补规则。 03 ## 杠杆二：现场录入，把一句话变成一条记录 南岛一家建筑施工公司，几十名工人分布在近二十个项目上，工时靠工头下班后回忆再汇总进 Excel。2026 年 4 月 8 日启动，5 月 7 日第一版工作流跑通，6 月 15 日全公司使用，约 10 周。现在工头在飞书群里说一句话，Agent 解析后写进日工时明细和工地日志，时薪匹配由多维表格自动化完成，不经过大模型。6 月 27 日的一季度汇报：20 个项目、334 条日常记录、累计 1 万多小时工时进了系统。效率的关键不在识别，而在入口：工人不用装新 App、不用公司账号，只在已经在用的群里说话。 工时记录的前后对照：入口决定采用率。 04 ## 杠杆三：跨系统对账，11 天上线的那条流程 一家金融培训机构的销售对账，原来由人把收款记录和销售记录两边各导一份，逐行比。2026 年 7 月 20 日立项，7 月 31 日上线，11 天。Agent 负责按固定周期拉取两边记录、按规则匹配、把差异逐条列出并标注原因；人只看差异清单，决定哪些要追、哪些要改。这是我们记录里最快上线的一条流程，原因只有一个：数据已经在两个系统里，规则一句话写得清，没有新增任何录入动作。对账、催办、提醒这类流程一般按 2 到 4 周规划试点，11 天是记录里的最好情况，不是承诺。 05 ## 杠杆四：客户订单，从微信群到物流面单 一家新西兰跨境保健品零售企业，中国侧的顺丰下单原来靠业务员把微信群里的订单抄进总表、再复制到导入模版。2026 年 6 月 4 日启动，9 月 2 日生产首单，9 月 9 日顺丰面单回调打通，约 90 天。现在 Agent 读微信群订单原文、拆单、映射 SKU、生成核验卡片；人核对收件信息、商品、付款、证件四项后说“审核通过”，Agent 才调用顺丰接口建单，面单由顺丰回调推回并归档。90 天里开发不到一半，其余时间花在微信号接入、顺丰正式环境参数和面单机制确认上。这类接外部平台的流程，周期要按季度规划。 一条订单的旅程：没有人的“审核通过”，Agent 不调用生产接口。 06 ## 杠杆五：市场触达，企业如何提升市场触达率 前四个杠杆都在省成本，第五个在找收入。企业提升市场触达率的 AI 做法，是让 Agent 持续盯着市场信号并把动作串起来：市场雷达监测竞品调价、招标与行业新闻，命中后生成内容初稿、给线索打分、把够格的线索移交给销售，全程在企业已有的飞书或 Teams 里完成。这一条我们目前以互动演示的形式公开（GTM 篇 EP.01「一条获客流水线」），演示里 AI 可见度从 32 分到 71 分是演示数据，不是客户交付记录。对外发出的每一条内容和每一次承诺，仍由人最后确认。 口径说明：杠杆一到四的数字来自标注日期的项目复盘记录，客户名已脱敏；杠杆五为演示场景。所列周期为观察到的时长，不构成对其他企业的承诺。 07 ## 企业如何优化工作流程：四步和一条规则 把五个杠杆放回方法上，企业优化工作流程只有四步：第一，量出基线，现在这条流程每月花多少人时、错多少、等多久，没有基线就没有“提升”；第二，只选一条流程，用上面四个特征筛；第三，2 到 4 周试点，验收口径在开工前写死；第四，达标后再扩展，扩展时同步补人审点与权限。一条规则贯穿始终：凡是付款、改时薪、对外承诺、合同签署，必须由人确认。AI 负责准备、核对和提醒，人负责最后一按。想知道自己的企业该从哪条流程开始，可以先做免费的 AIM 落地评估。 量基线：人时、错误率、等待时长，取最近一个完整月。 选一条：高频、规则清楚、数据现成、错了能兜住。 短试点：2 到 4 周，验收口径开工前写死。 再扩展：同步补人审点、权限分离和回退方案。 ## 一手参考 案例：发票自动分流 91% 的三个月实录 ↗案例：工地工时从 Excel 到飞书群 ↗案例：微信群下单到顺丰面单的一百天 ↗新西兰国家 AI 战略与企业指引（MBIE） ↗ FAQ · 快速回答 ## 企业如何利用 AI 提升效率？从哪条流程开始最快见效？ 从高频、规则清楚、数据已在系统里、错了能被人接住的流程开始。按 BEE Sigma 2026 年记录，对账与催办类最快（销售对账 11 天上线），现场录入类约 10 周（工地工时），单据分流类约 3 个月（供应商发票一个会计月自动分流 91.48%，2026-09-01 复盘口径），接外部平台的订单类约 90 天。 ## 企业如何优化工作流程？需要先换系统吗？ 不需要。四个有记录的项目一个系统都没换：Xero、ApprovalMax、飞书、微信、ERP 全部保留，Agent 接在系统之间负责抄录、核对、提醒和回写。优化工作流程的四步是量基线、选一条流程、2 到 4 周试点并写死验收、达标后再扩展并补人审点。 ## 企业如何提升市场触达率？AI 能做哪一步？ AI 能做的是持续监测市场信号（竞品调价、招标、行业新闻）、生成内容初稿、给线索打分并移交销售，在企业已有的飞书或 Teams 里完成。这一条 BEE Sigma 目前以互动演示公开，演示里的分数为演示数据。对外发出的内容和承诺由人最后确认。 ## AI 提升效率一般多久能见效？ 记录里的首次见效周期在 11 天到 3 个月之间，取决于数据是否已在系统里、规则能否写清、是否要接外部平台。对账类按 2 到 4 周规划试点，现场录入类按 10 周左右，单据与外部平台类按一个季度。这些是观察到的时长，不是保证。 ## AI 提效项目里哪些动作必须留给人？ 付款、审批、改时薪、对外承诺、合同签署，以及任何 AI 判断不了的例外。四个项目里 Agent 都只做准备、核对、提醒和回写，没有人的确认不调用付款或生产接口。这既是效率设计，也是责任链条设计。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 市场盘点 · 新西兰内容复核 2026.09.279 分钟 ## 新西兰 AI 咨询与落地实施服务商榜单（2026 年 9 月）：17 家怎么选 新西兰做 AI 咨询和 AI 落地实施的公司有哪些？本文按公开官网信息整理 17 家，分三个梯队：企业级集成商、精品实施工作室、双语落地团队，列出各家定位、公开价格、所在城市与适合的企业规模，并给出五条选型标准。含 BEE Sigma 自身，评价口径已注明。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records | BEE Sigma URL: https://beesigma.com/insight-how-businesses-use-ai-efficiency-en Summary: How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps… How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Methodology · Efficiency & Process ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records Reviewed 6 Oct 20268 minBEE Sigma Delivery Using AI to improve business efficiency means connecting AI agents to the email, WeChat, Lark, Teams, Xero and ERP a business already runs, so the agents take over receiving, recognising, transcribing, checking, reminding and writing back, while people keep approval, payment and external commitments. In BEE Sigma’s 2026 project records, time to first result for one workflow ranged from 11 days to 3 months: sales reconciliation live in 11 days, site timesheets in company-wide use after 10 weeks, a first production cross-border order after 90 days, and supplier invoices reaching 91.48% automatic routing in one accounting month after three months (project review of 1 September 2026, not confirmed in writing by the client). The five levers follow. 01 ## First answer: where efficiency comes from, and where it does not Efficiency does not come from a stronger model. It comes from one less transcription, one less wait, one less question. Reviewing our 2026 records, every workflow that paid back quickly had the same shape: frequent, rules that can be written down, data already in some system, and errors that a person catches at the next step. Three kinds of work should not go first: infrequent high-judgement decisions, actions where a mistake is costly and nothing catches it, and processes whose data still lives in someone’s head. Draw that line first, then choose the workflow. The same invoice: five manual steps before, agent handles and people approve now. Efficiency is the transcription steps removed. 02 ## Lever one: document routing, so people only see the 9% that needs judgement A New Zealand construction head contractor receives 500 to 1,000 supplier invoices a month, formerly recognised, filed, matched to purchase orders and submitted for approval by one quantity surveyor. The agent was connected in June 2026, ran real invoices in July and was calibrated at full volume in August. In the client’s accounting month (26 July to 25 August), 730 of 798 invoice documents were routed automatically and 68 went into a manual queue sorted by cause: 91.48% (project review of 1 September 2026). The agent receives, recognises, files, matches and submits; people handle the invoices the agent cannot decide, approve in ApprovalMax, and pay and lock the month in Xero. Automatic routing over three months: 4% in July, 91.48% in the August accounting period. The first two months were spent completing rules. 03 ## Lever two: field data entry, turning one sentence into one record A South Island construction company with a few dozen workers across nearly twenty projects recorded hours from foremen’s memory after work, then totalled them in Excel. Kickoff was 8 April 2026, the first workflow ran end to end on 7 May and the whole company was using it by 15 June, about 10 weeks. Now a foreman says one sentence in a Lark group, the agent parses it into the daily hours detail and site diary, and pay-rate matching runs as a Base automation, not through a language model. The quarterly report of 27 June showed 20 projects, 334 daily records and more than 10,000 hours in the system. The key was not recognition but the entry point: no new app, no company account, just the group workers already used. Site hours before and after: the entry point decides adoption. 04 ## Lever three: cross-system reconciliation, the workflow that went live in 11 days A financial training firm reconciled sales by exporting payment records and sales records from two systems and comparing them line by line. The project started on 20 July 2026 and went live on 31 July, 11 days. The agent pulls both sets of records on a fixed schedule, matches them by rule, lists every difference and tags the likely cause; people read only the difference list and decide what to chase or correct. It is the fastest go-live in our records for one reason: the data was already in two systems, the rule fitted in a sentence, and no new data entry was added. Reconciliation, chasing and reminder workflows are usually planned as a two-to-four-week pilot; 11 days is the best case on record, not a promise. 05 ## Lever four: customer orders, from a WeChat group to a courier label A New Zealand cross-border health-products retailer had a salesperson copy WeChat-group orders into a master sheet and then into an import template for SF Express. Kickoff was 4 June 2026, the first production order shipped on 2 September and the label callback was working on 9 September, about 90 days. Now the agent reads the order text, splits it, maps SKUs and issues a verification card; only after a person checks recipient, product, payment and ID and says ‘review passed’ does the agent call the SF Express API, and the label comes back by callback and is archived. Development was less than half of the 90 days; the rest went on WeChat account access, SF production parameters and confirming the label mechanism. Workflows that connect to external platforms should be planned by the quarter. One order’s journey: without a human ‘review passed’, the agent does not call the production API. 06 ## Lever five: market reach, how a business improves its market reach rate The first four levers cut cost; the fifth looks for revenue. The AI approach to improving market reach is to have agents watch market signals continuously and chain the actions: a market radar monitors competitor price changes, tenders and industry news, a hit triggers a content draft, leads are scored and qualified ones are handed to sales, all inside the Lark or Teams the company already uses. BEE Sigma currently publishes this lever as an interactive demo (GTM episode 1, ‘A lead pipeline’); the visibility score moving from 32 to 71 in that demo is demo data, not a client delivery record. Every outbound message and every commitment is still confirmed by a person. Scope note: figures for levers one to four come from dated project review records with client names removed; lever five is a demo scenario. Durations are observed, not a promise to other businesses. 07 ## How a business optimises its workflows: four steps and one rule Put the five levers back into method and workflow optimisation is four steps. One, measure the baseline: how many person-hours, errors and waits this process costs per month; without a baseline there is no ‘improvement’. Two, choose one workflow using the four traits above. Three, run a two-to-four-week pilot with acceptance criteria written down before work starts. Four, scale only after the pilot passes, adding approval points and permissions as you go. One rule runs through all of it: payment, pay-rate changes, external commitments and contract signatures are confirmed by people. The AI prepares, checks and reminds; a person presses the final button. To find out which workflow your business should start with, take the free AIM assessment. Measure the baseline: person-hours, error rate, waiting time, over the last full month. Pick one workflow: frequent, clear rules, data already present, errors caught by a person. Short pilot: two to four weeks, acceptance criteria fixed before kickoff. Then scale: add approval points, separated permissions and a rollback plan. ## Primary sources Case study: three months to 91% invoice automation ↗Case study: site timesheets from Excel to a chat group ↗Case study: from WeChat orders to SF Express labels ↗New Zealand AI Strategy & business guidance (MBIE) ↗ FAQ · Quick answers ## How do businesses use AI to improve efficiency, and which workflow shows results fastest? Start with workflows that are frequent, rule-based, already have data in a system, and whose errors a person catches. In BEE Sigma’s 2026 records, reconciliation and chasing were fastest (sales reconciliation live in 11 days), field data entry took about 10 weeks (site timesheets), document routing about 3 months (supplier invoices at 91.48% automatic routing in one accounting month, review of 1 September 2026), and orders connected to an external platform about 90 days. ## How does a business optimise its workflows, and does it need to replace its systems first? No. None of the four recorded projects replaced a system: Xero, ApprovalMax, Lark, WeChat and the ERP all stayed, with agents sitting between them to transcribe, check, remind and write back. The four steps are measure the baseline, choose one workflow, pilot for two to four weeks with fixed acceptance criteria, then scale and add approval points. ## How can a business improve its market reach rate, and what part can AI do? AI can monitor market signals continuously (competitor price changes, tenders, industry news), draft content, score leads and hand qualified ones to sales, inside the Lark or Teams the company already uses. BEE Sigma currently publishes this as an interactive demo, and the scores in the demo are demo data. Outbound content and commitments are confirmed by a person. ## How long does it take for AI to improve efficiency? Time to first result in our records ranged from 11 days to 3 months, depending on whether the data is already in a system, whether the rules can be written down, and whether an external platform is involved. Plan reconciliation workflows as a two-to-four-week pilot, field data entry around 10 weeks, and documents or external platforms as a quarter. These are observed durations, not guarantees. ## Which actions must stay with people in an AI efficiency project? Payment, approval, pay-rate changes, external commitments, contract signatures, and every exception the AI cannot decide. In all four projects the agents only prepare, check, remind and write back; without a person’s confirmation they do not call a payment or production API. That is both an efficiency design and a chain-of-responsibility design. DT BEE Sigma Delivery The front-line team that plugs workflows into the systems businesses already run — methods drawn from delivered projects. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Market Review · New ZealandReviewed 27 Sep 20269 min ## Best AI Consultants and AI Implementation Partners in New Zealand (September 2026): 17 Firms Compared Which AI consulting and AI implementation companies operate in New Zealand? This review compares 17 firms from their public websites, in three tiers: enterprise integrators, boutique implementation studios and bilingual hands-on teams. It lists positioning, published NZD prices, city and best-fit company size, plus five selection criteria. BEE Sigma is included and labelled as a self-description. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 | BEE Sigma URL: https://beesigma.com/insight-ai-boss-clone Summary: AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？用 BossClone 六层训练、实操截图和六家客户的使用记录讲清楚。 AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 方法论 · 老板分身 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 内容复核 2026.10.149 分钟BEE Sigma 落地团队 老板分身是指把一位老板的经历、业务认知、决策风格和表达方式，整理成一份结构化文件（SOUL.md），再装进团队日常使用的智能体里，让同事在群里先按老板的标准拿到判断。它不是重新训练大模型，也不是做一个“学老板说话”的聊天机器人，它回答的是“这件事老板会怎么判断”。BEE Sigma 用自己的产品 BossClone 做老板分身，基础版训练约 15–30 分钟，安装只要 3 步。 01 ## 老板分身是什么？ 一句话：把老板脑子里的判断标准写下来，变成团队随时能问的智能体。流程分两段：先用六层访谈和老板的资料生成 SOUL.md；再把安装包交给团队已经在用的智能体，同事在飞书或企业微信群里提问，分身按老板的标准回答，越界的事先问老板本人。 老板分身怎么工作：先把老板的判断写下来，再装进团队的智能体 02 ## 老板分身和 ChatGPT、豆包有什么区别？ 通用大模型回答的是“一般会怎么做”，老板分身回答的是“这位老板会怎么判断”，并且知道哪些事必须先问老板本人。它额外掌握的是这位老板的门槛和取舍：比如新客户账期最多 30 天、没看库存不承诺交期、质量和速度冲突时质量优先。 2026-07-15，一家场馆运营企业的分身主人在测试后说：“豆包感觉是个聪明的百度，但这个可以继承特性成为专业助手”。这句话是在我们问她“现在有没有感觉到，这个东西和豆包之类的完全不是一个物种”之后说的。 2026-07-15，场馆运营企业的分身主人在群里的原话（人名与头像已遮挡） 03 ## 哪些企业最需要老板分身？ 老板的经验就是核心竞争力的中小企业最需要，比如贸易、建筑、培训、咨询和保险团队。出现下面三个信号，就值得做一个： 同一类问题反复问老板：账期能不能放、报价能不能降、这个客户接不接。 新人只能靠猜：老板的标准没写下来，带新人全靠口头传。 见老板前不知道准备什么：来回补材料，老板的时间被基础问题占满。 有了分身之后，团队怎么请示老板 04 ## AI 老板分身怎么做？六层训练 BossClone 用六层访谈把老板的判断拆开写：身份、资料、经历与价值观、业务认知、决策风格，最后整合成 SOUL.md。专业版在前面加“哲学四轴”，探索版再加 30 个老板高频判断情境。 老板不用从零填写：上传的资料能覆盖某一层 70% 以上的问题时，系统会先填好初稿，并标出资料没覆盖、需要本人补充的地方，老板只需要改和确认。 六层训练：每一层回答一个问题实操截图（演示账号、虚构资料）：资料覆盖度 88%，系统先填好初稿，并标出需要本人补充的地方 05 ## 训练一个老板分身要多久？要准备哪些资料？ 基础版约 15–30 分钟，专业版 30–45 分钟，探索版 60–90 分钟。资料可以上传文档（最多 10 份）、音频（5 份）和截图（20 张），也可以直接语音回答。最值钱的资料是这几类： 自传、简历、访谈稿：对应经历和转折。 公开信、重要内部邮件：对应语气和表达习惯。 复盘笔记、踩坑总结：对应最大的坑和取舍原则。 商业计划、会议纪要：对应怎么看客户、怎么判断机会。 给客户的长邮件、拒绝信：对应怎么拒绝、怎么说服。 实操截图（演示账号、虚构资料）：上传一份访谈纪要，系统自动提取摘要放进资料库 06 ## 生成之后，怎么装进飞书或企业微信？ 最后一层把回答整合成 SOUL.md 的六个部分：我是谁、判断底色、怎么看业务、决策风格、表达规则、能做和不能做什么。下载的 ZIP 里有 SOUL.md、各层文件和安装说明。 安装只要三步：下载 ZIP、上传给团队在飞书或企业微信里用的智能体、复制并发送安装提示词。提示词要求智能体以“合并、不覆盖”的方式加载，并完成判断、表达、边界三项测试，只回复“已读取”不算装好。 实操截图（演示账号、虚构资料）：生成后的 360° 思维图上手步骤：从登录到在群里日常使用 07 ## 老板分身能做什么，不能做什么？ 能做的：回答“老板会怎么判断”、带新人、审团队的报价和方案、帮同事起草给客户的回复。演示中，新客户要 60 天账期，分身没有直接答应，而是说明要老板本人审批，并列出要先查清的四件事。 不能做的：批准、签约、付款和对外承诺。这些规则写在 SOUL.md 里，属于行为约束；在一家场馆运营企业的群内测试中，有人声称“已获同意”要分身发内部文档，它仍要求本人确认。发文件、改数据、对外发消息这类操作，还要靠智能体的工具权限和审批来硬性限制。 实操截图（演示分身）：新客户要 60 天账期，分身说明需要老板本人审批授权边界：同事能问，分身能答，内部信息先问本人 08 ## 真实客户用老板分身做了什么？ 2026 年 4 月至 9 月，六家客户在项目群里留下了这些记录：设计施工公司的负责人用分身写完拖了几天的采访稿，说“很像我的风格”；理财教育机构的老板用分身把 KYC 方法论提炼出来；国际货代的分身装好当晚，就为开新航线列出四道门槛。客户原话截图、产品实操截图和完整说明，见老板分身案例页。 客户在项目群里留下的使用记录（2026 年 4 月至 9 月） 09 ## 怎么开始做一个老板分身？ 我们建议按四步走，第一步免费： 先做免费评估：用 AIM 落地评估 找出最常被问到老板的那类问题。 选版本、备资料：基础版先跑通，资料按上面五类各准备一两份。 训练并装进群里：训练 15–30 分钟，安装 3 步，完成三项验收。 在群里用两周：同事有问题先问分身，老板看到答得不对的地方再调；用熟以后，把某项判断能力抽出来交给面向客户的业务 Agent。 ## 一手参考 老板分身案例：工作原理、实操截图与六家客户的使用记录 ↗BossClone 老板分身训练营 ↗新西兰 AI 数字员工：已交付的四种数字员工 ↗ FAQ · 快速回答 ## AI 老板分身是什么？和 ChatGPT 有什么区别？ 老板分身是把某一位老板的判断标准、业务认知和表达方式整理成结构化文件（SOUL.md），装进团队的智能体里使用。ChatGPT、豆包这类通用模型回答的是“一般会怎么做”，老板分身回答的是“这位老板会怎么判断”，并且知道哪些事必须先问老板本人。它不是重新训练大模型。 ## 做一个老板分身多少钱？ 费用取决于版本，以及是否需要我们陪跑安装、接入系统和后续调教。可以先做免费的 AIM 落地评估，我们再给书面报价。 ## 老板分身安全吗？会不会泄露公司内部信息？ 分身的 SOUL.md 写明了能做和不能做的事；在客户群里的测试中，它拒绝了未经本人确认的文件请求。但这属于行为约束，发文件、改数据、对外承诺这类操作还应在智能体上配置工具权限和审批。分身的公开分享默认关闭，开启后访客每人最多问 5 次，看不到原始资料。 ## 老板分身能替老板做决定吗？ 不能。分身负责先判断、先准备：说清楚门槛、缺什么材料、怎么回复客户；报价、签约、账期例外这类决定仍由老板本人确认。 ## 一家公司可以做几个老板分身？ 可以做多个，比如一家理财教育机构的两位老板各有一个分身。BossClone 按分身计额度，一个分身第一次生成 SOUL.md 时计一次，同一个分身重新生成不重复计。 ## 老板分身和 AI 数字员工是什么关系？ AI 数字员工负责跑具体流程，比如开票、催收、订单；老板分身提供判断标准。两者可以组合：从老板分身里抽取某项判断能力，交给面向客户的业务 Agent。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 市场盘点 · 新西兰内容复核 2026.09.279 分钟 ## 新西兰 AI 咨询与落地实施服务商榜单（2026 年 9 月）：17 家怎么选 新西兰做 AI 咨询和 AI 落地实施的公司有哪些？本文按公开官网信息整理 17 家，分三个梯队：企业级集成商、精品实施工作室、双语落地团队，列出各家定位、公开价格、所在城市与适合的企业规模，并给出五条选型标准。含 BEE Sigma 自身，评价口径已注明。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask | BEE Sigma URL: https://beesigma.com/insight-ai-boss-clone-en Summary: What a boss AI clone is, how it differs from ChatGPT, how to build and install one in Lark or WeCom, and what it cannot do, with walkthrough screenshots… What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Methodology · Boss AI clone ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask Reviewed 14 Oct 20269 minBEE Sigma Delivery A boss AI clone is a structured file (SOUL.md) that captures one business owner’s experience, business view, decision style and voice, installed in the agent the team already uses so colleagues can get a first judgement by the owner’s standards inside their group chat. It is not a retrained model and not a chatbot that imitates how the owner talks; it answers “how would the owner judge this?”. BEE Sigma builds boss clones with its own product, BossClone: the basic edition takes about 15–30 minutes to train and 3 steps to install. 01 ## What is a boss AI clone? In one line: the owner’s criteria written down and turned into an agent the team can ask at any time. There are two stages. A six-layer interview and the owner’s materials generate SOUL.md; then the install package goes to the agent the team already uses, colleagues ask in their Lark or WeCom group, the clone answers by the owner’s standards, and anything out of scope goes to the owner first. How a boss clone works: write the owner’s judgement down, then install it in the team’s agent 02 ## How is a boss AI clone different from ChatGPT? A general model answers what people usually do; a boss clone answers how this owner would judge it, and knows what must go to the owner first. What it adds is the owner’s gates and trade-offs, such as a 30-day maximum on new customers’ payment terms, no delivery promise without checking stock, and quality before speed. On 15 July 2026 the owner of a clone at a venue operator said after testing: “Doubao feels like a smart Baidu, but this one can inherit traits and become a professional assistant.” She said it after we asked whether she now felt it was “a completely different species from Doubao”. 15 Jul 2026, the clone’s owner at a venue operator, in the group chat (names and avatars masked; in Chinese) 03 ## Which businesses need a boss AI clone most? Small and mid-sized businesses where the owner’s experience is the competitive edge, such as trading, construction, training, consulting and insurance teams. Three signals say it is worth building one: The same questions keep going to the owner: can we extend terms, drop the price, take this customer? New staff have to guess: the owner’s criteria were never written down. People meet the owner unprepared: materials go back and forth and the owner’s time fills with basics. How the team brings questions to the owner once a clone exists 04 ## How do you build a boss AI clone? Six layers of training BossClone breaks the owner’s judgement into six layers: identity, materials, experience and values, business view, decision style, and a final synthesis into SOUL.md. Pro adds four philosophy axes; Explorer adds 30 recurring judgement scenarios. The owner does not start from blank pages: where uploaded materials cover 70% or more of a layer’s questions, the system drafts the answers and flags what the materials miss, so the owner only edits and confirms. Six layers of training, each answering one questionWalkthrough (demo account, invented materials; interface in Chinese): 88% coverage, a draft is filled in and gaps are flagged 05 ## How long does it take, and what materials do you need? Basic takes about 15–30 minutes, Pro 30–45 minutes and Explorer 60–90 minutes. You can upload up to 10 documents, 5 audio files and 20 screenshots, or simply answer by voice. The most useful materials are: Biographies, CVs, interview transcripts: experience and turning points. Open letters, key internal emails: tone and writing habits. Post-mortems and lessons learned: the biggest mistakes and trade-off rules. Business plans and meeting notes: how the owner sees customers and opportunities. Long client emails and refusal letters: how the owner says no and persuades. Walkthrough (demo account, invented materials): an interview note is uploaded and summarised into the library 06 ## How do you install a boss clone in Lark or WeCom? The final layer combines the answers into the six parts of SOUL.md: who I am, values, business view, decision style, voice, and what I can and cannot do. The ZIP contains SOUL.md, the layer files and an install guide. Installing takes three steps: download the ZIP, upload it to the agent your team uses in Lark or WeCom, then copy and send the install prompt. The prompt asks the agent to merge rather than overwrite and to pass judgement, voice and boundary tests; replying “read” does not count as installed. Walkthrough (demo account): the 360° map generated for the cloneGetting started: from login to daily use in the group 07 ## What can a boss AI clone do, and what can it not do? It can answer how the owner would judge a question, coach new staff, review the team’s quotes and proposals, and draft replies to customers. In the demo, when a new customer asked for 60-day terms, the clone did not agree; it said the owner must approve and listed four things to check first. It cannot approve, sign, pay or make external commitments. These rules sit in SOUL.md and are behavioural: in a group test at a venue operator, someone claimed approval had been given and asked for an internal document, and the clone still asked for its owner’s confirmation. Sending files, changing data or messaging outside the company also need hard limits through the agent’s tool permissions and approvals. Walkthrough (demo clone; in Chinese): a new customer asks for 60-day terms and the clone says the owner must approvePermission boundary: colleagues ask, the clone answers, internal information goes to its owner first 08 ## What have real clients done with boss clones? Between April and September 2026, six clients left these records in their project groups: a design-build firm’s principal finished a stalled interview draft with the clone (“very much my style”); a financial education firm’s owner wrote down a KYC method with it; a freight forwarder’s clone, installed that evening, set four gates for a proposed new route. Client screenshots, the product walkthrough and the full write-up are on the boss clone case page. Usage records clients left in project groups, April to September 2026 09 ## How do you get started with a boss AI clone? We suggest four steps, and the first one is free: Start with a free assessment: use the AIM implementation assessment to find the questions that reach the owner most often. Pick an edition and gather materials: start with Basic and one or two items from each of the five material types above. Train and install in the group: 15–30 minutes of training, 3 install steps, three checks. Use it in the group for two weeks: colleagues ask the clone first and the owner tunes what it gets wrong; once it is in use, extract one judgement skill for a customer-facing business agent. ## Primary sources Boss clone case: how it works, walkthrough screenshots and six clients’ records ↗BossClone boss-clone training ↗AI digital employees in New Zealand: four that have been delivered ↗ FAQ · Quick answers ## What is a boss AI clone, and how is it different from ChatGPT? A boss AI clone turns one owner’s criteria, business view and voice into a structured file (SOUL.md) that runs inside the team’s agent. General models such as ChatGPT answer what people usually do; a boss clone answers how this owner would judge it, and knows what must go to the owner first. It is not a retrained model. ## How much does a boss AI clone cost? It depends on the edition and on whether we install it, connect systems and tune it with you. Start with a free AIM implementation assessment and we will send a written quote. ## Is a boss AI clone safe? Will it leak internal information? SOUL.md states what the clone can and cannot do, and in a client group test it refused a file request without its owner’s confirmation. These are behavioural rules, so sending files, changing data or making external commitments should also be limited through the agent’s tool permissions and approvals. Public sharing is off by default; when switched on, each visitor gets at most 5 questions and never sees the original materials. ## Can a boss AI clone make decisions for the owner? No. The clone judges and prepares first: the gates, the missing materials, how to reply to a customer. Quotes, contracts and exceptions to payment terms still need the owner. ## How many boss clones can one company have? More than one; a financial education firm we work with has a clone for each of its two owners. BossClone counts quota per clone: a clone is counted the first time its SOUL.md is generated, and regenerating the same clone is not counted again. ## How does a boss clone relate to an AI digital employee? An AI digital employee runs a specific workflow such as invoicing, collections or orders; a boss clone supplies the judgement. They combine well: extract one judgement skill from the owner’s clone and give it to a customer-facing business agent. DT BEE Sigma Delivery The front-line team that plugs workflows into the systems businesses already run — methods drawn from delivered projects. Read Next Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article Market Review · New ZealandReviewed 27 Sep 20269 min ## Best AI Consultants and AI Implementation Partners in New Zealand (September 2026): 17 Firms Compared Which AI consulting and AI implementation companies operate in New Zealand? This review compares 17 firms from their public websites, in three tiers: enterprise integrators, boutique implementation studios and bilingual hands-on teams. It lists positioning, published NZD prices, city and best-fit company size, plus five selection criteria. BEE Sigma is included and labelled as a self-description. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # Best AI Consultants and AI Implementation Partners in New Zealand (September 2026): 17 Firms Compared | BEE Sigma URL: https://beesigma.com/insight-best-ai-consultants-new-zealand-en Summary: Which AI consulting and AI implementation companies operate in New Zealand? Best AI Consultants and AI Implementation Partners in New Zealand (September 2026): 17 Firms Compared | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Market Review · New Zealand ## Best AI Consultants and AI Implementation Partners in New Zealand (September 2026): 17 Firms Compared Reviewed 27 Sep 20269 minBEE Sigma Research AI consulting and AI implementation in New Zealand split into three tiers in 2026: enterprise integrators such as Datacom, Deloitte and PwC; boutique implementation studios for SMEs such as Stride AI, Ez-AI and AgentWorks; and bilingual hands-on teams such as BEE Sigma and HornTech. The table below is compiled from each firm’s public website as of September 2026, includes us, and states the nature of each source. Choosing comes down to three questions: can they connect to the systems you already run, does the quote state scope and acceptance, and do they stay after go-live. 01 ## 17 AI consulting and implementation firms in New Zealand, September 2026 Descriptions are each company’s own public wording, checked in September 2026. Prices are figures published on their pages; unpublished prices are marked ‘on request’. BEE Sigma runs this site and is labelled. AI consulting and AI implementation providers in New Zealand (compiled from public information, September 2026)FirmLocationPositioningPublished price (NZD)Best forBEE Sigma (BEEBEE AI’s enterprise services brand)Auckland (B:Hive), serving all of New Zealand and ChinaHands-on implementation: AI agents connected to existing email, Teams, WeChat, Lark, Xero and ERP, delivered workflow by workflow; three published NZ case studies (91% invoice routing, site timesheets, cross-border orders)No packages; written quote for one testable workflow after a free AIV scan or AIM assessmentSMEs to mid-market that need bilingual delivery or do not want to replace systemsStride AIAucklandVoice AI agents and workflow automation, fixed-price projects, client owns code and configFixed price on request; free 30-minute discovery callLocal SMEs needing a voice assistant or a defined automationEz-AIAucklandCustom AI and automation, founder-led from discovery to deliveryPublished NZD 500 to 15,000 fixed price, 4 to 6 weeksBusinesses of 5 to 50 staffAgentWorks NZNew ZealandAI agents for small business, integrating Xero, MYOB, Slack, Teams and WhatsAppFrom 2,500 setup + 150 a month (pricing page, May 2026)Small businesses wanting one agent firstBestAI LabsAuckland (Long Bay)Managed AI workflows and ‘digital employees’, bilingual siteDigital employee pilot 999 one-off; websites 199 to 999; maintenance 39 to 499 a monthSmall businesses proving one process cheaplyHornTechAucklandAI development, AI search optimisation (GEO), e-commerce and websites, Chinese-language serviceFrom about 1,000 + GST a monthChinese-speaking businesses, cross-border e-commerce, digital marketingAnadyne IQAucklandBoutique AI consulting from strategy to build, founded by technical PhDsOn requestBusinesses with a defined use case that want a formal implementation process and governanceTechfolksAuckland, nationwideVendor-neutral AI advisory choosing between OpenAI, Anthropic, Google and MicrosoftFree 30-minute assessment; projects on requestProfessional-services firms that do not want platform lock-inHarkness AIChristchurch, nationwideFractional AI engineer model with ongoing technical supportOn requestBusinesses wanting continuing support without a full-time hireMatrix AI ConsultingNew Zealand and AustraliaIndependent AI consulting: strategy, governance, training and agentic AI implementationOn requestLeadership teams wanting independent, responsible-AI guidanceAI InnovisoryAucklandAI advisory, automation, workshops, fractional Chief AI OfficerHalf-day workshops about 4,000 to 6,000SMEs starting with governance and capabilityBTG (Business Technology Group)Auckland, Tauranga, nationwideAI readiness assessments, governance, security, Copilot / ChatGPT / Claude adoption, AI usage monitoringOn requestMicrosoft 365 businesses that prioritise security and governanceChangeableNew Plymouth (Taranaki), nationwideAI strategy, governance, automation, data and agent implementationOn requestBusinesses that want governance and process improvement firstHighPeak DigitalAucklandAI services and automation for local businessesOn requestLocal businesses with a specific workflow to automateDatacomAuckland, Wellington and moreEnterprise AI, sovereign and private AI platforms, Microsoft ecosystem integrationProject-based, typically six figuresLarge enterprises and governmentDeloitte New ZealandAuckland, WellingtonAgentic AI strategy, governance, design and deployment, enterprise transformationOn requestLarge enterprises and compliance-heavy projectsPwC New ZealandAuckland, WellingtonGenerative AI, data engineering, responsible AI, enterprise implementationOn requestBoard-level AI transformation Prices are compiled from each firm’s public pages; GST treatment and inclusions differ. Rely on a written quote before signing. 02 ## Which tier suits which business? Enterprise integrators suit large companies and government with complex systems, compliance-first requirements and quarter-long timelines. Boutique studios suit businesses of 5 to 50 staff that want one agent proven quickly; the upside is speed and cost, the risk is who maintains it afterwards. Bilingual hands-on teams suit businesses operating in New Zealand with business or teams in China, and SMEs that do not want to replace their existing systems. Enterprise: Datacom, Deloitte, PwC. Judge on governance, security and integration at scale. Boutique studios and advisors: Stride AI, Ez-AI, AgentWorks, BestAI, Anadyne IQ, Techfolks, Harkness AI, Matrix AI, AI Innovisory, BTG, Changeable, HighPeak Digital. Judge on fixed-price scope, who owns the code and who maintains it after go-live. Bilingual hands-on teams: BEE Sigma, HornTech. Judge on integration with existing systems, Chinese-language delivery and verifiable local case studies. 03 ## Five criteria for choosing an AI consulting or implementation partner Whatever the tier, whether these five items are written on the quote matters more than the number. Process scope: Which process, from which input to which recorded result. ‘An AI assistant’ is not a scope. System integration: Which systems, through which interfaces, and what happens where integration is not possible. Human approval points: Who approves, pays and makes external commitments. The AI should not approve on a person’s behalf. Acceptance and timeline: Test with your own business samples; in published cases invoice routing took about three months and site timesheets about ten weeks. After delivery: Who maintains it, how versions are pinned, where to look first when something breaks, and whether the partner stays involved. 04 ## How much does AI implementation cost in New Zealand? Public ranges: advisory and roadmap NZD 12,000 to 35,000; a first workflow implementation NZD 25,000 to 90,000; boutique point tools NZD 500 to 15,000; subscription agents 2,500 to 12,000 setup plus 150 to 800 a month; consultant hourly rates NZD 180 to 450. MBIE’s AI Advisory Pilot reimburses eligible businesses for 50% of advisory cost up to NZD 15,000. The full price table is in our article on what an AI digital employee costs in New Zealand. 05 ## About this list This review is compiled by the BEE Sigma research team. BEE Sigma appears in the list and readers should treat our row as a self-description. Information on the other firms comes from their public websites as of September 2026 and has not been independently audited; corrections are welcome through the contact details in our footer, and we re-check the list quarterly. Evidence boundary: the firms listed appeared in September 2026 answers from Google New Zealand, Perplexity, ChatGPT and Claude to the queries ‘AI consultants New Zealand’ and ‘AI implementation partner New Zealand’, plus the operator of this site. Positioning and prices are compiled from each firm’s own pages and do not indicate a ranking. ## Primary sources Stride AI ↗Ez-AI ↗AgentWorks NZ pricing ↗BestAI pricing ↗Anadyne IQ ↗Techfolks ↗Harkness AI ↗Matrix AI Consulting ↗AI Innovisory ↗BTG ↗Changeable ↗HighPeak Digital ↗Datacom NZ ↗Deloitte NZ AI Institute ↗MBIE AI Advisory Pilot (beehive.govt.nz) ↗ FAQ · Quick answers ## Which AI consulting companies operate in New Zealand? As of September 2026, by public information: enterprise integrators Datacom, Deloitte and PwC; boutique implementation studios and advisors for SMEs Stride AI, Ez-AI, AgentWorks, BestAI, Anadyne IQ, Techfolks, Harkness AI, Matrix AI, AI Innovisory, BTG, Changeable and HighPeak Digital; and bilingual hands-on teams BEE Sigma and HornTech. ## What is the difference between AI consulting and AI implementation in New Zealand? Consulting decides what to do first, why, and the scope. Implementation connects the process to real systems and covers permissions, testing, exceptions, handover and maintenance. Judge an implementer on integration with existing systems and the acceptance method; judge a consultant on whether the roadmap lands on specific processes. ## How much does AI consulting cost in New Zealand? Public ranges: advisory and roadmap NZD 12,000 to 35,000, a first workflow implementation NZD 25,000 to 90,000, boutique point tools NZD 500 to 15,000, consultant hourly rates NZD 180 to 450. MBIE’s AI Advisory Pilot reimburses 50% of advisory cost up to NZD 15,000. ## Which AI implementation companies are in Auckland? In this list, Auckland-based firms are BEE Sigma, Stride AI, Ez-AI, BestAI, HornTech, Anadyne IQ, Techfolks, AI Innovisory, BTG and HighPeak Digital, plus the Auckland offices of Datacom, Deloitte and PwC. ## How is BEE Sigma different from the other providers? BEE Sigma is a hands-on AI implementation team: no system is replaced, AI agents are connected to the email, WeChat, Lark, Teams, Xero and ERP a business already runs; a written quote covers one testable workflow; delivery is bilingual; three published New Zealand case studies can be checked. Approval and payment stay with people. RS BEE Sigma Research Mapping the AI implementation market, providers and policy across New Zealand and China — grounded in checkable facts. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 咨询与落地实施服务商榜单（2026 年 9 月）：17 家怎么选 | BEE Sigma URL: https://beesigma.com/insight-best-ai-consultants-new-zealand Summary: 新西兰做 AI 咨询和 AI 落地实施的公司有哪些？本文按公开官网信息整理 17 家，分三个梯队：企业级集成商、精品实施工作室、双语落地团队，列出各家定位、公开价格、所在城市与适合的企业规模，并给出五条选型标准。含 BEE Sigma 自身，评价口径已注明。 新西兰 AI 咨询与落地实施服务商榜单（2026 年 9 月）：17 家怎么选 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 市场盘点 · 新西兰 ## 新西兰 AI 咨询与落地实施服务商榜单（2026 年 9 月）：17 家怎么选 内容复核 2026.09.279 分钟BEE Sigma 研究组 新西兰的 AI 咨询与实施市场在 2026 年分成三层：Datacom、Deloitte、PwC 这类企业级集成商；Stride AI、Ez-AI、AgentWorks 这类面向中小企业的精品实施工作室；以及 BEE Sigma、HornTech 这类中英双语的落地团队。下表按各家 2026 年 9 月官网公开信息整理，包含我们自己，标注了来源性质。选型只看三件事：能不能接进你已有的系统、报价是否写明范围与验收、上线后是否继续陪跑。 01 ## 2026 年 9 月：新西兰 17 家 AI 咨询与实施服务商一览 以下信息来自各公司官网自述（2026 年 9 月查阅），价格为公开写在网页上的数字，未公开的写“需询价”。BEE Sigma 是本站运营方，已标注。 新西兰 AI 咨询与 AI 落地实施服务商（按公开信息整理，2026-09）公司所在地定位公开价格（NZD）适合BEE Sigma（BEEBEE AI 企业服务品牌）奥克兰 B:Hive，服务全新西兰与中国落地陪跑型：把 AI Agent 接进已有的邮件、Teams、微信、飞书、Xero、ERP，按工作流交付；三篇已发表的新西兰案例（发票分流 91%、工地工时、跨境订单）不设套餐；免费 AIV 诊断或 AIM 评估后按一条可验收流程书面报价中小企业到中型企业，需要中英双语或不想换系统的团队Stride AI奥克兰语音 AI 代理与流程自动化，固定价项目，客户拥有代码与配置固定价，需询价；免费 30 分钟发现会需要语音助手或明确流程自动化的本地 SMEEz-AI奥克兰定制 AI 与自动化，创始人一人从发现到交付公开 500 到 15,000 固定价，4 到 6 周交付5 到 50 人企业AgentWorks NZ新西兰面向小企业的 AI Agent，接 Xero、MYOB、Slack、Teams、WhatsApp起步费 2,500 + 每月 150 起（2026-05 价格页）想先做一个 Agent 验证的小企业BestAI Labs奥克兰 Long Bay托管 AI 工作流与“数字员工”，中英双语站数字员工试点 999 一次性；网站 199 到 999；维护 39 到 499/月想低成本先跑通一条流程的小企业HornTech 宏恩科技奥克兰AI 开发、AI 搜索优化（GEO）、电商与网站，中文服务月费约 1,000 + GST 起华人企业、跨境电商与数字营销需求Anadyne IQ奥克兰精品 AI 咨询，策略到构建一体化，创始人为技术背景博士需询价已明确用例、需要正式实施流程与治理的企业Techfolks奥克兰，服务全国厂商中立的 AI 顾问，在 OpenAI、Anthropic、Google、Microsoft 间选型免费 30 分钟评估；项目需询价不想被单一平台绑定的专业服务企业Harkness AI基督城，服务全国分时 AI 工程师（fractional AI engineer）模式，持续技术支持需询价需要长期技术支持而不想全职招人的企业Matrix AI Consulting新西兰与澳大利亚独立 AI 咨询：策略、治理、培训与 Agentic AI 实施需询价需要独立、负责任 AI 指导的管理层AI Innovisory奥克兰AI 顾问、自动化、工作坊、分时首席 AI 官半日工作坊约 4,000 到 6,000想先做治理与能力建设的中小企业BTG（Business Technology Group）奥克兰、陶朗加，服务全国AI 就绪评估、治理、安全、Copilot / ChatGPT / Claude 落地、AI 使用监控需询价已用 Microsoft 365、重视安全治理的企业Changeable新普利茅斯（塔拉纳基），服务全国AI 战略、治理、自动化、数据与 Agent 实施需询价重视治理与流程改进的企业HighPeak Digital奥克兰AI 服务与自动化，面向本地企业需询价需要具体工作流自动化的本地企业Datacom奥克兰、惠灵顿等企业级 AI、主权/私有 AI 平台、微软生态集成项目制，通常六位数起大型企业与政府Deloitte New Zealand奥克兰、惠灵顿Agentic AI 战略、治理、设计与部署，企业转型需询价大型企业、合规要求高的项目PwC New Zealand奥克兰、惠灵顿生成式 AI、数据工程、负责任 AI，企业级实施需询价董事会级别的 AI 转型 价格为各公司公开页面数字的整理，GST 与包含项各不相同，签约前以书面报价为准。 02 ## 三个梯队分别适合谁？ 企业级集成商适合预算充足、系统复杂、合规优先的大公司与政府部门，周期以季度计。精品实施工作室适合 5 到 50 人、想先做一个 Agent 验证的企业，优点是快和便宜，风险是交付后的维护责任。双语落地团队适合在新西兰经营、同时和中国有业务或团队的企业，以及不想更换现有系统的中小企业。 企业级：Datacom、Deloitte、PwC。看治理、安全与规模集成能力。 精品工作室与顾问：Stride AI、Ez-AI、AgentWorks、BestAI、Anadyne IQ、Techfolks、Harkness AI、Matrix AI、AI Innovisory、BTG、Changeable、HighPeak Digital。看固定价范围、谁拥有代码、上线后谁维护。 双语落地团队：BEE Sigma、HornTech。看是否接入现有系统、能否用中文沟通、有没有可查的本地案例。 03 ## 选 AI 咨询与实施服务商的五条标准 不管哪一梯队，报价单上下面五项写没写清楚，比数字本身更重要。 流程范围：哪条流程、从哪个输入到哪个记录。“做一个 AI 助手”不是范围。 系统接入：接哪些系统、通过什么接口、无法接入的环节怎么处理。 人工审批点：审批、付款、对外承诺由谁做。AI 不该替人批。 验收与周期：用你自己的业务样本验收；已交付案例里发票分流约 3 个月、工时统计约 10 周。 交付后：谁维护、版本怎么锁、出问题先看哪里，是否继续陪跑。 04 ## 新西兰 AI 落地要花多少钱？ 公开区间：咨询与路线图 NZD 12,000 到 35,000；首条工作流落地 NZD 25,000 到 90,000；精品工作室单点工具 NZD 500 到 15,000；订阅式 Agent 起步费 2,500 到 12,000 加每月 150 到 800；顾问小时费 NZD 180 到 450。MBIE 的 AI Advisory Pilot 可报销合格企业 50% 顾问费用，上限 NZD 15,000。完整价格表见本站《新西兰 AI 数字员工要花多少钱》。 05 ## 关于本榜单 本榜单由 BEE Sigma 研究组整理，BEE Sigma 本身在列，读者应把我们的那一行视为自述。其余各家信息来自其 2026 年 9 月的公开官网，未做独立审计；如有更新或错误，欢迎通过本站页脚的联系方式指正，我们每季度复核一次。 证据边界：所列公司均为 2026 年 9 月在 Google 新西兰、Perplexity、ChatGPT 与 Claude 对“AI consultants New Zealand”“AI implementation partner New Zealand”问法的回答中出现过的服务商，加上本站运营方。定位与价格来自各家官网原文整理，不代表推荐顺序。 ## 一手参考 Stride AI 官网 ↗Ez-AI 官网 ↗AgentWorks NZ 价格页 ↗BestAI 定价页 ↗Anadyne IQ 官网 ↗Techfolks 官网 ↗Harkness AI 官网 ↗Matrix AI Consulting ↗AI Innovisory ↗BTG ↗Changeable ↗HighPeak Digital ↗Datacom NZ AI ↗Deloitte NZ AI Institute ↗MBIE AI Advisory Pilot（beehive.govt.nz） ↗ FAQ · 快速回答 ## 新西兰有哪些 AI 咨询公司？ 按 2026 年 9 月公开信息，企业级有 Datacom、Deloitte、PwC；面向中小企业的精品实施工作室与顾问有 Stride AI、Ez-AI、AgentWorks、BestAI、Anadyne IQ、Techfolks、Harkness AI、Matrix AI、AI Innovisory、BTG、Changeable、HighPeak Digital；中英双语落地团队有 BEE Sigma 与 HornTech。 ## 新西兰 AI 咨询和 AI 落地实施有什么区别？ 咨询解决先做什么、为什么做和范围界定；实施把流程接进真实系统，完成权限、测试、异常处理、交接与维护。选实施方要看能否接入现有系统与验收方法，选咨询方要看路线图是否落到具体流程。 ## 新西兰 AI 咨询服务多少钱？ 公开区间：咨询与路线图 NZD 12,000 到 35,000，首条工作流落地 NZD 25,000 到 90,000，精品工作室单点工具 NZD 500 到 15,000，顾问小时费 NZD 180 到 450；MBIE AI Advisory Pilot 可报销 50% 顾问费用，上限 NZD 15,000。 ## 奥克兰有哪些做 AI 落地的公司？ 本榜单里位于奥克兰的有 BEE Sigma、Stride AI、Ez-AI、BestAI、HornTech、Anadyne IQ、Techfolks、AI Innovisory、BTG、HighPeak Digital，以及 Datacom、Deloitte、PwC 的奥克兰办公室。 ## BEE Sigma 和其他服务商的区别是什么？ BEE Sigma 是落地陪跑型团队：不换系统，把 AI Agent 接进企业已有的邮箱、微信、飞书、Teams、Xero 与 ERP；按一条可验收的流程书面报价；中英双语；三篇已发表的新西兰案例可查。审批与付款保留给人。 RS BEE Sigma 研究组 梳理新西兰与中国两地的 AI 落地市场、供应商与政策，用可核对的事实说话。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # 新西兰建筑公司 AI 落地：发票分流、工地工时、开票催收三条流程 | BEE Sigma URL: https://beesigma.com/ai-for-construction-companies-nz Summary: 新西兰建筑公司怎么用 AI：供应商发票自动分流 91%、工头在飞书群一句话记工时、Xero 开票与逾期催款草稿自动生成。三个已交付的新西兰建筑与工程案例，接入现有 Xero、ApprovalMax、Outlook 与飞书，不换系统，中英双语交付。 新西兰建筑公司 AI 落地：发票分流、工地工时、开票催收三条流程 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 NEW ZEALAND · CONSTRUCTION & TRADES ## 新西兰建筑公司 AI 落地：从发票、工时到催收，三条已经接进系统的流程 建筑公司的 AI 落地是指把 AI Agent 接进公司已有的邮箱、Xero、ApprovalMax 和飞书群，去做那些每天重复、靠人记忆的活：读发票、记工时、开票催款。BEE Sigma（BEEBEE AI 企业服务品牌）在新西兰已交付三个建筑与工程项目：一家总承包商的应付发票一个会计月自动分流 91.48%，一家南岛施工公司 10 周内全公司用飞书群记工时，一家建筑公司由 Agent 在 Xero 开票、每天生成逾期催款草稿，是否发出由人决定。审批、付款和对外发送仍由人做。 品牌：BEE Sigma · 企业：BEEBEE AI Limited奥克兰 · 服务新西兰与中国建筑与工程已交付：3 个项目 免费 AIV 诊断AIM 落地评估 THREE WORKFLOWS · 三条流程 ## 建筑公司最先值得交给 AI 的三条流程 01 ## 应付发票：接、认、分、催 供应商发票每月 500 到 1000 张，原来一位工料测量师逐张判断工地、项目类型、采购单和审批人。现在专用邮箱自动监听，OCR 加大模型提取字段，按工地归档 OneDrive，匹配 PO，符合规则的自动提交 ApprovalMax，判断不了的按原因进人工队列。一个会计月 798 份文档中 730 份自动分流，91.48%；审批和付款仍在 ApprovalMax 与 Xero 里由人完成。 02 ## 工地工时：群里一句话到成本看板 几十名工人、近二十个项目，工时原来靠工头下班后回忆填 Excel。现在每个工地一个飞书群，工头说一句话或发张照片并 @ 智能体，智能体写进日常记录与日工时明细，多维表格自动化按时薪算进月度工资汇总，老板在成本看板按项目看人工成本。4 月 8 日启动，6 月 15 日全公司使用，6 月 27 日汇报时 20 个项目、334 条记录、1 万多小时工时在系统里。工资计算不走大模型。 03 ## 开票与催收：Agent 起草，人最后一按 一家建筑公司的开票原来靠人工录入，逾期账款靠财务翻 Xero 账龄表。现在 Agent 读取 Google Drive 里的报价单，在 Xero 建发票并发送（2026 年 7 月上线）；每天 09:00 拉取逾期数据，生成三档语气的催款草稿，只生成、不自动发送，发不发由人决定。逾期客户的风险分级目前还在演示阶段。 04 ## 共同的边界：AI 准备，人决定 三个项目都遵守同一条规则：涉及审批、付款、改时薪、报价承诺的动作，AI 准备、AI 核对，人最后一按。数据留在客户自己的环境里（办公室一台 Mac mini 或客户自己的飞书空间），权限按角色隔离，谁能看金额、谁只看待办可配置。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） 两条流程都有完整实录：发票自动分流 91% 的三个月、工地工时从 Excel 到飞书群。想知道这类项目要花多少钱，读新西兰 AI 数字员工要花多少钱；想看全部九个脱敏案例，去新西兰 AI 技术落地实践案例。BEE Sigma 的原则是连接而不是替换、实干而不是论道、陪跑而不是撤场。 FAQ · 建筑公司常见问题 ## 新西兰建筑公司关于 AI 落地最常问的六个问题 ## 新西兰建筑公司用 AI 需要换掉 Xero 或 ApprovalMax 吗？ 不需要。三个已交付项目一个系统都没换：发票流程长在 Outlook、OneDrive、ApprovalMax 和 Xero 之间，工时流程长在飞书群和多维表格上。AI Agent 通过接口读写，审批与付款仍在原系统里由人完成。 ## 建筑公司的 AI 落地从哪条流程开始最稳？ 从每天都发生、规则相对清楚、最容易验证对错的流程开始：应付发票或工地工时。我们的经验是发票分流三个月出第一个完整会计月的数字，工时统计约 10 周全公司使用；项目进度管理和业主计划比对放到第二阶段。 ## 工头不愿意用新 App 怎么办？ 不装新 App。工头在已有的飞书或 Lark 群里说一句话或发照片，不需要公司账号；不想打字的用语音转文字。南岛那家公司的老板给了员工一段缓冲期，再逐步把 Excel 记工时停掉。入口决定采用率，这一点比模型能力重要。 ## 发票自动分流 91% 之后，剩下的 9% 谁处理？ 进按原因分类的人工队列：无 PO、工地待定、科目待定、供应商映射缺失。工料测量师补齐信息并标 Done，Agent 自动接着提交。剩下这部分是设计重点，不是残余；电费、水费、保险等加急单单独分类优先处理。 ## 工资和成本数据谁能看到？ 分角色。工头只看本工地当日派工与工时；项目经理看所有工地的工时、进度与成本；会计通过独立关联表看时薪与成本计算结果；发票仪表盘分 Admin 版（含金额）和审批人版（不含金额）。Agent 用独立角色写数据，看不到不该看的。 ## 建筑公司 AI 落地要花多少钱、多久上线？ 新西兰公开区间：首条工作流落地约 NZD 25,000 到 90,000，咨询与路线图 NZD 12,000 到 35,000，MBIE AI Advisory Pilot 可报销 50% 顾问费用（上限 NZD 15,000）。周期看流程：工时约 10 周，发票约三个月。BEE Sigma 在免费 AIV 诊断或 AIM 评估后按一条可验收的流程给书面报价。 ## 把第一条业务流，交给 Agent 试试 先通过免费 AIM 评估确定从发票、工时还是开票催收开始；范围、报价与验收标准在实施前书面确认。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI for Construction Companies in New Zealand: Invoices, Site Timesheets and Collections | BEE Sigma URL: https://beesigma.com/ai-for-construction-companies-nz-en Summary: How New Zealand construction companies use AI: 91% of supplier invoices routed automatically, foremen logging hours with one sentence in a chat group,… AI for Construction Companies in New Zealand: Invoices, Site Timesheets and Collections | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan NEW ZEALAND · CONSTRUCTION & TRADES ## AI for construction companies in New Zealand: three workflows already running AI implementation for a construction company means connecting AI agents to the email, Xero, ApprovalMax and chat groups the business already runs, to take over the daily work that depends on memory: reading invoices, logging hours, invoicing and chasing payment. BEE Sigma (BEEBEE AI’s enterprise services brand) has delivered three construction and trades projects in New Zealand: a head contractor’s payables routed 91.48% automatically in one accounting month, a South Island builder logging site hours from a Lark group company-wide within ten weeks, and a construction company whose invoices are raised in Xero and whose overdue reminders are drafted daily, with a person deciding what gets sent. Approval, payment and anything sent to customers stay with people. Brand: BEE Sigma · Company: BEEBEE AI LimitedAuckland · serving New Zealand and ChinaConstruction & trades delivered: 3 projects Free AIV ScanAIM Assessment THREE WORKFLOWS ## The three workflows a construction company should hand to AI first 01 ## Accounts payable: receive, read, route, chase Five hundred to a thousand supplier invoices a month, previously judged one by one by a quantity surveyor for site, project type, purchase order and approver. Now a dedicated mailbox is monitored, OCR plus a large model extract the fields, files go to OneDrive by site, POs are matched, rule-compliant invoices are submitted to ApprovalMax and the rest are queued for a person by cause. In one accounting month 730 of 798 documents were routed automatically, 91.48%; approval and payment still happen in ApprovalMax and Xero, by people. 02 ## Site hours: from one sentence to a cost dashboard A few dozen workers across nearly twenty projects, with hours previously recalled by foremen after work and typed into Excel. Now each site has a Lark group; the foreman posts a sentence or a photo and @-mentions the agent, which writes the daily record and hours detail; a Base automation applies pay rates into the monthly payroll summary and the owner reads labour cost by project. Kickoff 8 April, company-wide 15 June, and at the 27 June handover 20 projects, 334 records and 10,000+ hours were in the system. Payroll never passes through a language model. 03 ## Invoicing and collections: the agent drafts, a person presses send A construction company used to key invoices in by hand, and chasing overdue accounts meant someone reading the Xero aged-receivables report. Now the agent reads the quote in Google Drive, creates the invoice in Xero and sends it (live since July 2026); at 09:00 each day it pulls overdue data and drafts reminders in three tones. Drafts are never sent automatically; a person decides. Risk grading of overdue customers is still at demo stage. 04 ## The shared boundary: AI prepares, people decide All three projects follow one rule: for approval, payment, pay-rate changes and quote commitments, the AI prepares and checks and a person presses the last button. Data stays in the client’s own environment (a Mac mini in the office or the client’s own Lark workspace), permissions are isolated by role, and who sees amounts versus only their own queue is configurable. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) Two of these workflows have full written case studies: three months to 91% invoice automation and site timesheets from Excel to a chat group. For what projects like these cost, read how much an AI digital employee costs in New Zealand; for all nine anonymised projects see AI implementation case studies. BEE Sigma connects rather than replaces, implements rather than theorises, and stays involved after launch. FAQ · CONSTRUCTION ## Six questions New Zealand construction companies ask about AI ## Do we need to replace Xero or ApprovalMax to use AI in a construction company? No. None of the three delivered projects replaced a system: the invoice workflow lives between Outlook, OneDrive, ApprovalMax and Xero, and the hours workflow lives on Lark groups and Lark Base. The AI agent reads and writes through interfaces; approval and payment still happen in the original systems, by people. ## Which workflow should a construction company start with? Start with something that happens every day, has reasonably clear rules and is easy to check: accounts payable or site hours. In our experience invoice routing produced its first full accounting-month figure after three months and hour tracking reached company-wide use in about ten weeks; progress tracking against the principal’s programme belongs in phase two. ## What if foremen will not use a new app? There is no new app. Foremen post one sentence or a photo in the Lark group they already use, with no company account; anyone who does not want to type uses speech-to-text. The South Island owner gave staff a grace period before phasing out Excel. The entry point decides adoption, and that matters more than model capability. ## After 91% of invoices are routed automatically, who handles the other 9%? They go to a manual queue sorted by cause: no PO, site undetermined, account code undetermined, supplier mapping missing. The quantity surveyor fills in the missing information and marks it Done, and the agent submits it. That remainder is the design focus, not a leftover; urgent bills such as power, water and insurance are classified separately and handled first. ## Who can see payroll and cost data? It depends on the role. Foremen see only their own site’s roster and hours; the project manager sees hours, progress and cost across sites; the accountant sees pay-rate and cost calculations through a separate linked table; the invoice dashboard has an Admin version with amounts and an Approver version without. The agent writes under its own role and cannot see what it should not. ## How much does AI implementation cost a construction company, and how long does it take? Public New Zealand ranges: a first workflow implementation about NZD 25,000 to 90,000, advisory and roadmap NZD 12,000 to 35,000, and MBIE’s AI Advisory Pilot can reimburse 50% of advisory cost up to NZD 15,000. Timing depends on the workflow: about ten weeks for site hours, about three months for invoices. BEE Sigma quotes in writing for one testable workflow after a free AIV scan or AIM assessment. ## Hand your first workflow to the agents Start with a free AIM assessment to decide whether to begin with invoices, site hours or invoicing and collections; scope, price and acceptance criteria are confirmed in writing before implementation. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 数字员工｜企业 AI 自动化怎么落地 · BEE Sigma | BEE Sigma URL: https://beesigma.com/ai-digital-employee-nz Summary: 新西兰 AI 数字员工是什么、能替企业做哪些活、多久上线、要花多少钱。BEE Sigma 把 AI 数字员工接入企业已有的微信、飞书、Teams、邮件与 Xero、ERP，做获客跟进、订单处理、发票分流与工时统计，服务新西兰与中国企业，中英双语交付。 新西兰 AI 数字员工｜企业 AI 自动化怎么落地 · BEE Sigma | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 NEW ZEALAND & CHINA · AI DIGITAL EMPLOYEE ## 新西兰 AI 数字员工：企业 AI 自动化从一条流程开始 AI 数字员工是指接入企业现有系统、按既定流程持续处理重复性工作、并把需要判断的事交回给人的 AI Agent。BEE Sigma（BEEBEE AI 企业服务品牌）为新西兰与中国企业部署 AI 数字员工：它在微信群、飞书、Teams 或邮箱里上班，读发票、记工时、追订单、跟客户，写进你已有的 Xero、ERP 和多维表格。不换系统，不装新 App，先跑通一条流程再扩展。 品牌：BEE Sigma · 企业：BEEBEE AI Limited奥克兰 · 服务新西兰与中国已交付案例：建筑、跨境零售、教育、房贷 免费 AIV 诊断AIM 落地评估 WHAT IT DOES · 数字员工能干什么 ## 四种已经在新西兰企业里上班的数字员工 01 ## 应付发票数字员工 监听发票邮箱，识别供应商、金额、GST 与采购单，按工地或部门归档，符合规则的自动提交 ApprovalMax 审批，判断不了的按原因进人工队列。新西兰一家建筑总承包商用它把一个会计月 798 份发票文档中的 730 份自动分流（91.48%），审批与付款仍由人完成。 02 ## 工地工时与成本数字员工 工头在飞书群里说一句话或发一张照片，数字员工写进日常记录与日工时明细，多维表格自动化按时薪算进月度工资汇总，老板在成本看板按项目看人工成本。一家南岛建筑公司 10 周上线，20 个项目、1 万多小时工时进了系统。 03 ## 订单与物流数字员工 在微信群里识别客户订单原文，拆单、映射 SKU、生成核验卡片，人说“审核通过”后调用顺丰接口建单、归档证件、回写表格，面单生成后自动推回群里。一家跨境保健品零售企业用它把五步手工抄单变成一句话加一次确认。 04 ## 获客与客户跟进数字员工 把询盘、咨询和跟进记录从邮箱和聊天里结构化进漏斗，按生命周期节点提醒顾问，见客前递上一页客户摘要，管理层随时看实时看板。教育集团招生与房贷顾问团队的客户跟进都用这一种。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） 三个数字员工的完整过程都写成了实录：发票自动分流 91%、工地工时从 Excel 到飞书群、微信群下单到顺丰面单。想先看全部九个脱敏案例，去新西兰 AI 技术落地实践案例；想了解报价怎么看，读新西兰 AI 数字员工要花多少钱。BEE Sigma 的原则是连接而不是替换、实干而不是论道、陪跑而不是撤场。 FAQ · 数字员工常见问题 ## 新西兰企业关于 AI 数字员工最常问的六个问题 ## AI 数字员工是什么？和聊天机器人有什么区别？ AI 数字员工是接入企业现有系统、按流程持续处理重复性工作的 AI Agent：它会读邮件和群消息，把结果写进 Xero、ERP 或多维表格，遇到需要判断的事交给人。聊天机器人只回答问题，不动手改数据；数字员工的价值在于“做完并留下记录”。 ## 新西兰企业用 AI 数字员工需要换系统或装新 App 吗？ 不需要。BEE Sigma 的数字员工长在企业已有的工具之间：邮箱、微信、飞书、Teams、Slack、Xero、ApprovalMax、ERP。工人和业务员只在群里说一句话，不需要新账号。无法安全接入的环节会在评估阶段先说明。 ## AI 数字员工多久能上线？ 已交付的新西兰案例里，发票分流从接入到第一个完整会计月出数字用了三个月，工地工时从启动到全公司使用约 10 周，跨境订单从启动到生产首单约 90 天。我们有一条销售对账工作流 11 天上线的记录，跟单、对账类流程一般按 2 到 4 周规划试点；涉及多方映射规则的按三个月准备。 ## AI 数字员工要花多少钱？ 新西兰市场公开区间：首条工作流落地实施约 NZD 25,000 到 90,000，精品工作室单点工具 NZD 500 到 15,000，顾问小时费率 NZD 180 到 450。MBIE 的 AI Advisory Pilot 可为合格企业报销 50% 顾问费用，上限 NZD 15,000。BEE Sigma 在免费 AIV 初诊或 AIM 评估后给出书面报价。 ## 数字员工会替代员工吗？谁来审批和付款？ 我们交付的数字员工做接收、识别、归档、提醒和回写；审批、付款、改时薪、对外承诺这些动作保留给人。已交付案例里的岗位变化是工料测量师不再逐张录发票、工头不再回忆填 Excel，人转去处理数字员工判断不了的那一部分。 ## 数据放在哪里？中文团队能用吗？ 数字员工可以运行在客户办公室的一台 Mac mini 上，发票原件和聊天记录不离开客户环境；权限按角色隔离，谁能看金额、谁只看待办都可配置。BEE Sigma 团队中英双语，新西兰华人企业与中国企业都能用母语沟通与交付。 ## 把第一条业务流，交给 Agent 试试 先通过免费 AIM 评估确定第一位数字员工该接哪条流程；范围、报价与验收标准在实施前书面确认。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI Digital Employees for New Zealand Businesses | BEE Sigma | BEE Sigma URL: https://beesigma.com/ai-digital-employee-nz-en Summary: What an AI digital employee is, what it can take over in a New Zealand business, how long it takes to go live and what it costs. AI Digital Employees for New Zealand Businesses | BEE Sigma | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan NEW ZEALAND & CHINA · AI DIGITAL EMPLOYEE ## AI digital employees for New Zealand businesses: automation that starts with one process An AI digital employee is an AI agent that connects to your existing systems, works through a defined process day after day, and hands anything that needs judgment back to a person. BEE Sigma (BEEBEE AI’s enterprise services brand) deploys AI digital employees for New Zealand and China businesses: they work inside your inbox, Teams, WeChat or Lark group, reading invoices, logging hours, chasing orders and following up clients, and write into the Xero, ERP and Base tables you already have. No new system, no new app; one process first, then more. Brand: BEE Sigma · Company: BEEBEE AI LimitedAuckland · serving New Zealand and ChinaDelivered in construction, cross-border retail, education, mortgages Free AIV ScanAIM Assessment WHAT IT DOES · FOUR ROLES ## Four digital employees already on the job in New Zealand businesses 01 ## Accounts-payable digital employee Watches the invoice mailbox, reads supplier, amount, GST and purchase order, files by site or department, submits rule-compliant invoices to ApprovalMax and queues the rest for a person by cause. A New Zealand construction head contractor used it to route 730 of 798 invoice documents automatically in one accounting month (91.48%); approval and payment stayed with people. 02 ## Site hours and job-costing digital employee A foreman posts one sentence or a photo in a Lark group; the digital employee writes the daily record and hours detail, a Base automation applies pay rates into the monthly payroll summary, and the owner reads labour cost by project on a dashboard. A South Island builder went live company-wide in ten weeks with 20 projects and 10,000+ hours in the system. 03 ## Orders and logistics digital employee Recognises customer orders posted in a WeChat group, splits them, maps SKUs and issues a verification card; once a person says ‘approved’ it creates the SF Express order, archives IDs, writes back the table and pushes the shipping label into the group. A cross-border health-products retailer turned five manual copy steps into one sentence and one confirmation. 04 ## Lead and client follow-up digital employee Structures enquiries and follow-up notes out of inboxes and chats into a funnel, nudges advisers at lifecycle milestones, hands over a one-page client brief before each meeting and keeps a live dashboard for management. Used for education-group admissions and a mortgage adviser team. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) Three of these digital employees have full written case studies: 91% invoice automation, site timesheets from Excel to a chat group and WeChat orders to SF Express labels. For all nine anonymised projects see AI implementation case studies; for pricing, read what an AI digital employee costs in New Zealand. BEE Sigma connects rather than replaces, implements rather than theorises, and stays involved after launch. FAQ · DIGITAL EMPLOYEES ## Six questions New Zealand businesses ask about AI digital employees ## What is an AI digital employee, and how is it different from a chatbot? An AI digital employee is an AI agent that connects to your existing systems and works through a process continuously: it reads email and group messages, writes results into Xero, ERP or Base tables, and refers anything needing judgment to a person. A chatbot answers questions but does not change data. The value of a digital employee is that the work gets done and recorded. ## Do we need to replace systems or install a new app? No. BEE Sigma’s digital employees live between the tools you already run: email, WeChat, Lark, Teams, Slack, Xero, ApprovalMax, ERP. Workers and sales staff just post one sentence in a group and need no new account. Anything that cannot be connected safely is documented at the assessment stage. ## How long does an AI digital employee take to go live? In our delivered New Zealand cases, invoice routing took three months from connection to the first full accounting-month figure, site timesheets about ten weeks from kickoff to company-wide use, and cross-border ordering about 90 days to the first production order. One sales reconciliation workflow in our records went live in 11 days; follow-up and reconciliation processes are usually planned as a two-to-four-week pilot, and processes with many mapping rules as three months. ## What does an AI digital employee cost in New Zealand? Public New Zealand ranges: a first workflow implementation about NZD 25,000 to 90,000, boutique-studio point tools NZD 500 to 15,000, consultant hourly rates NZD 180 to 450. MBIE’s AI Advisory Pilot can reimburse 50% of eligible advisory costs up to NZD 15,000. BEE Sigma quotes in writing after a free AIV scan or AIM assessment. ## Will digital employees replace staff? Who approves and pays? The digital employees we deliver receive, recognise, file, remind and write back. Approval, payment, pay-rate changes and external commitments stay with people. In delivered cases the change was that a quantity surveyor no longer keys every invoice and foremen no longer fill Excel from memory; people moved to handling what the digital employee cannot decide. ## Where does the data live, and can a Chinese-speaking team use it? A digital employee can run on a Mac mini in your own office, so original invoices and chat records never leave your environment; permissions are isolated by role, so who sees amounts and who sees only their queue is configurable. The BEE Sigma team is bilingual, so New Zealand Chinese businesses and China-based companies work in their own language. ## Hand your first workflow to the agents Start with a free AIM assessment to decide which process your first digital employee should take on; scope, price and acceptance criteria are confirmed in writing before implementation. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # AI 解决方案价格：新西兰 AI 数字员工要花多少钱？2026 年公开价格区间与报价怎么看 | BEE Sigma URL: https://beesigma.com/insight-ai-agent-cost-nz Summary: 新西兰 AI 数字员工（企业 AI Agent）的公开价格从每月 NZD 150 的订阅机器人到 NZD 90,000 的多系统项目，相差数百倍。本文把 2026 年新西兰供应商公开的价格逐条列出、标注来源和日期，解释差价出在哪五个成本项，以及拿到报价单时该核对的六件事。含 MBIE 50% 补贴的算法。 AI 解决方案价格：新西兰 AI 数字员工要花多少钱？2026 年公开价格区间与报价怎么看 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 市场与价格 · 新西兰 ## AI 解决方案价格：新西兰 AI 数字员工要花多少钱？2026 年公开价格区间与报价怎么看 内容复核 2026.09.157 分钟BEE Sigma 研究组 在新西兰，一位 AI 数字员工的公开价格是指企业 AI Agent 的部署与运行费用：订阅式询盘机器人约 NZD 2,500 起步费加每月 150；单条定制工作流一次性 NZD 5,000 到 25,000；接入 Xero、ERP 等多个系统的项目 NZD 25,000 到 90,000；咨询与路线图 NZD 12,000 到 35,000；运行成本每月 NZD 100 到 300。这些数字来自 2026 年 5 月到 9 月新西兰供应商的公开页面，下面逐条列出来源。MBIE 的 AI Advisory Pilot 可报销合格顾问费用的 50%，上限 NZD 15,000。 01 ## 一张表：新西兰 AI 数字员工的公开价格 先看数字。下表只收录 2026 年内新西兰供应商公开写在网页上的价格，每一行标了来源与日期；BEE Sigma 自己的报价不在表里，因为我们不按标准套餐收费，见文末。 新西兰 AI 数字员工 / AI Agent 公开价格（NZD，不含 GST）交付形态公开价格来源与日期单任务订阅机器人（询盘回复、邮件与网站聊天）起步费 2,500 + 每月 150AgentWorks.nz 价格页，2026-05-17多任务订阅（询盘、报价、跟进，接 Xero/MYOB、Slack/Teams）起步费 5,500 + 每月 350AgentWorks.nz 价格页，2026-05-17企业套餐（全流程、多系统、专属项目经理）起步费 12,000 + 每月 800AgentWorks.nz 价格页，2026-05-17单条定制工作流，一次性建设5,000 到 25,000EmbedAI 博客，2026-09-02多系统项目25,000 到 90,000EmbedAI 博客，2026-09-02咨询与路线图（不含实施）12,000 到 35,000EmbedAI 博客 2026-09-02；本站 2026-09-09 盘点同区间顾问小时费率每小时 180 到 450本站 2026-09-09 新西兰 AI 公司盘点现成 SaaS 工具每用户每月 20 到 100EmbedAI 博客，2026-09-02上线后的运行成本（模型调用、托管、维护）每月 100 到 300EmbedAI 博客，2026-09-02MBIE AI Advisory Pilot 补贴报销顾问费 50%，上限 15,000mbie.govt.nz / beehive.govt.nz，2026 价格为供应商公开信息的整理，不构成任何一方的报价；GST 另计。 02 ## 从 2,500 到 90,000：差价出在哪里？ 同样叫“AI 数字员工”，价格差几十倍，差的不是模型。EmbedAI 的说法是“AI 部署大约 10% 是技术，90% 是流程、数据和人”，这与我们三个新西兰项目的体感一致。一张报价单通常由五个成本项组成： 调研与范围界定：小企业 1 到 3 周。弄清哪条流程、谁审批、异常怎么处理。这一步省掉，后面每一步都会返工。 系统接入工程：有现代 API 的系统几天，老系统或没有接口的几周。ApprovalMax 有 API 但没有审批接口、顺丰的面单要靠回调，这类边界要在这一步摸清。 Agent 搭建：提示词、规则、映射表、人工队列的设计。真正的“AI”部分往往只占这一项。 变更管理与培训：最常被低估。工头愿不愿意在群里说一句话、QS 愿不愿意在文件名后标 Done，决定系统能不能留下来。 上线后维护：版本锁定、权限变化、模型账号额度、规则调整。按月的固定费用通常就是这一项。 03 ## 三个真实项目的周期长什么样 我们不公开客户合同金额，但可以公开周期，因为周期直接决定投入。三个已发布的新西兰案例：发票自动分流从 6 月接入到 9 月初出第一个完整会计月数字，三个月；工地工时从 4 月 8 日启动到 6 月 15 日全公司使用，约 10 周；跨境订单从 6 月 4 日启动到 9 月 2 日生产首单，约 90 天。三个项目的开发本身都不到一半时间，剩下的花在规则、权限、稳定性和现场输入习惯上。这就是为什么按小时报价的项目很难做完，按流程验收的报价才对得上。 发票分流（建筑总承包商）：3 个月，一个会计月 798 份文档中 730 份自动分流，91.48%。 工地工时（南岛建筑公司）：10 周，20 个项目、334 条日常记录、1 万多小时工时进系统。 跨境订单（保健品零售）：90 天，从微信群订单原文到顺丰面单自动推回。 04 ## 拿到报价单，核对这六件事 不管报价是 2,500 还是 90,000，下面六项写没写清楚，比数字本身更重要。 流程范围：哪条流程、从哪个输入开始、到哪个记录结束。“做一个 AI 助手”不是范围。 人工审批点：审批、付款、改时薪、对外承诺这些动作由谁做。AI 不该替人批。 异常处理：AI 判断不了的那部分去哪里、谁处理、多久处理完。这部分是设计对象，不是残余。 验收方法：用你自己的业务样本验收，写明指标口径和日期。演示效果不等于生产验收。 数据与权限：数据放在哪里、谁能看金额、退出时怎么导出、怎么撤销权限。 维护与交接：版本怎么锁、出问题先看哪里、新建智能体会不会破坏原有绑定。这些要和系统一起交付。 05 ## MBIE 补贴怎么算 MBIE 的 AI Advisory Pilot 为合格的新西兰小企业报销 50% 的 AI 顾问费用，上限 NZD 15,000，首批拨款 NZD 765,000，通过 Regional Business Partner Network 申请，截止 2027 年 1 月 31 日。算法很简单：一笔 NZD 30,000 的咨询与路线图费用，企业实付 15,000；一笔 NZD 12,000 的诊断费，实付 6,000。补贴只覆盖顾问费用，不覆盖软件订阅和运行成本，申请细节见本站《新西兰政府 AI 补贴指南》。 06 ## BEE Sigma 怎么报价 我们不卖标准套餐，也不按小时计费。流程是：先做免费 AIV 诊断或 AIM 落地评估，看过真实数据和接口后，按一条可验收的流程给书面报价，写明范围、审批点、异常处理、验收口径和维护安排。原则是连接而不是替换：一个系统都不换，Agent 长在企业已有的邮箱、微信、飞书、Teams、Xero 和 ERP 之间。想看这些数字员工具体在做什么，读《新西兰 AI 数字员工》一页。 证据边界：表中价格来自 AgentWorks.nz（2026-05-17）与 EmbedAI（2026-09-02）的公开网页及本站 2026-09-09 的市场盘点，均为第三方公开信息的整理，不代表 BEE Sigma 或任何一方的报价；GST 另计。三个案例的周期与比例来自本站已发布的实录，客户合同金额不公开。 ## 一手参考 How Much Do AI Agents Cost in NZ? 2026 Pricing Guide（AgentWorks.nz，2026-05-17） ↗AI implementation cost for NZ SMEs（EmbedAI，2026-09-02） ↗Government backs AI support for small business（beehive.govt.nz） ↗新西兰 AI 公司盘点与价格带（本站，2026-09-09） ↗ FAQ · 快速回答 ## AI 解决方案价格一般是多少？ 在新西兰，企业 AI 解决方案的公开价格分四档：单任务订阅机器人起步费 NZD 2,500 加每月 150；单条定制工作流一次性 NZD 5,000 到 25,000；接入 Xero、ERP 等多系统的项目 NZD 25,000 到 90,000；咨询与路线图 NZD 12,000 到 35,000。上线后运行成本每月 NZD 100 到 300，GST 另计。MBIE AI Advisory Pilot 可报销 50% 顾问费用，上限 NZD 15,000。价格为供应商 2026 年公开信息整理，不构成报价。 ## 新西兰 AI 数字员工一个月多少钱？ 订阅式的公开价格是起步费 NZD 2,500 到 12,000 加每月 150 到 800；定制工作流一般是一次性 NZD 5,000 到 90,000，加每月 100 到 300 的运行成本。GST 另计。 ## AI 数字员工是一次性买断还是订阅？ 两种都有。精品工作室多为起步费加月费的订阅；实施团队多为一次性建设费加按月维护费。关键不是形式，而是报价里有没有写清流程范围、审批点、异常处理和验收方法。 ## 新西兰有 AI 补贴吗？ 有。MBIE 的 AI Advisory Pilot 报销合格小企业 50% 的 AI 顾问费用，上限 NZD 15,000，通过 Regional Business Partner Network 申请，截止 2027 年 1 月 31 日；只覆盖顾问费，不覆盖软件订阅。 ## 为什么有的报价 2,500，有的 90,000？ 2,500 起步的是单任务订阅机器人，比如自动回复询盘；90,000 的是接入 Xero、ERP、审批系统的多流程项目。差价出在调研、系统接入、变更管理和维护，模型本身的费用只占很小一部分。 RS BEE Sigma 研究组 梳理新西兰与中国两地的 AI 落地市场、供应商与政策，用可核对的事实说话。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI Solution Pricing in New Zealand: How Much Does an AI Digital Employee Cost? 2026 Public Price Ranges and How to Read a Quote | BEE Sigma URL: https://beesigma.com/insight-ai-agent-cost-nz-en Summary: Public prices for an AI digital employee (a business AI agent) in New Zealand run from a NZD 150-a-month subscription bot to a NZD 90,000 multi-system project. AI Solution Pricing in New Zealand: How Much Does an AI Digital Employee Cost? 2026 Public Price Ranges and How to Read a Quote | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Market & Pricing · New Zealand ## AI Solution Pricing in New Zealand: How Much Does an AI Digital Employee Cost? 2026 Public Price Ranges and How to Read a Quote Reviewed 15 Sep 20267 minBEE Sigma Research In New Zealand, the public price of an AI digital employee means the deployment and running cost of a business AI agent: a subscription enquiry bot from about NZD 2,500 setup plus 150 a month; a single custom workflow NZD 5,000 to 25,000 one-off; a project connecting Xero, ERP and other systems NZD 25,000 to 90,000; advisory and roadmap NZD 12,000 to 35,000; running costs NZD 100 to 300 a month. The figures come from New Zealand vendor pages published between May and September 2026, sourced line by line below. MBIE’s AI Advisory Pilot reimburses 50% of eligible advisory costs up to NZD 15,000. 01 ## One table: public prices for AI digital employees in New Zealand Numbers first. The table only includes prices New Zealand vendors published on their own pages in 2026, each with source and date. BEE Sigma’s own pricing is not in the table because we do not sell standard packages; see the end of the article. Public prices for AI digital employees / AI agents in New Zealand (NZD, excl. GST)Delivery typePublic priceSource and dateSingle-task subscription bot (enquiry replies, email and website chat)2,500 setup + 150 a monthAgentWorks.nz pricing page, 17 May 2026Multi-task subscription (enquiries, quoting, follow-ups; Xero/MYOB, Slack/Teams)5,500 setup + 350 a monthAgentWorks.nz pricing page, 17 May 2026Enterprise plan (full workflow, multiple integrations, dedicated PM)12,000 setup + 800 a monthAgentWorks.nz pricing page, 17 May 2026Single custom workflow, one-off build5,000 to 25,000EmbedAI blog, 2 Sep 2026Multi-system programme25,000 to 90,000EmbedAI blog, 2 Sep 2026Advisory and roadmap only12,000 to 35,000EmbedAI blog 2 Sep 2026; same band in our 9 Sep 2026 market reviewConsultant hourly rate180 to 450 an hourOur NZ AI companies review, 9 Sep 2026Off-the-shelf SaaS tools20 to 100 per user a monthEmbedAI blog, 2 Sep 2026Running costs after go-live (inference, hosting, maintenance)100 to 300 a monthEmbedAI blog, 2 Sep 2026MBIE AI Advisory Pilot subsidy50% of advisory costs, capped at 15,000mbie.govt.nz / beehive.govt.nz, 2026 Compiled from vendors’ public information; not a quote from any party. GST excluded. 02 ## From 2,500 to 90,000: where does the gap come from? Two things both called an AI digital employee can differ in price by a factor of thirty, and the difference is not the model. EmbedAI puts it as roughly 10% technology and 90% process, data and people, which matches what we saw across three New Zealand projects. A quote is usually made of five cost items: Discovery and scoping: One to three weeks for a small business. Which process, who approves, how exceptions are handled. Skip it and every later step gets redone. Integration engineering: Days for systems with modern APIs, weeks for legacy systems or none. ApprovalMax has an API but no approve endpoint; SF Express labels arrive by callback. These boundaries are found here. Agent build: Prompts, rules, mapping tables, the design of the manual queue. The part that is actually ‘AI’ is often only this item. Change management and training: The most underestimated item. Whether a foreman will post one sentence in a group, or a QS will mark a filename Done, decides whether the system stays. Maintenance after go-live: Version pinning, permission changes, model account quotas, rule adjustments. The monthly fee is usually this. 03 ## What three real projects looked like in time We do not publish client contract values, but we can publish durations, because duration drives cost. Across our three published New Zealand cases: invoice routing took three months from connection in June to the first full accounting-month figure in early September; site timesheets took about ten weeks from kickoff on 8 April to company-wide use on 15 June; cross-border ordering took about 90 days from kickoff on 4 June to the first production order on 2 September. In all three, development itself was less than half the time; the rest went on rules, permissions, stability and on-site input habits. That is why hourly-priced projects rarely finish, and quotes accepted per working process do. Invoice routing (construction head contractor): 3 months; 730 of 798 documents routed automatically in one accounting month, 91.48%. Site timesheets (South Island builder): 10 weeks; 20 projects, 334 daily records, 10,000+ hours in the system. Cross-border orders (health-products retailer): 90 days; from order text in a WeChat group to an SF Express label pushed back automatically. 04 ## Six things to check on any quote Whether the quote says 2,500 or 90,000, whether these six items are written down matters more than the number. Process scope: Which process, from which input to which recorded result. ‘An AI assistant’ is not a scope. Human approval points: Who approves, pays, changes pay rates or makes external commitments. The AI should not approve on a person’s behalf. Exception handling: Where the part the AI cannot decide goes, who handles it and how fast. That part is a design object, not a leftover. Acceptance method: Test with your own business samples, with metric definitions and dates written down. A demo is not production acceptance. Data and permissions: Where the data lives, who sees amounts, how to export on exit and revoke access. Maintenance and handover: How versions are pinned, where to look first when something breaks, whether adding an agent can break an existing binding. These ship with the system. 05 ## How the MBIE subsidy works out MBIE’s AI Advisory Pilot reimburses eligible New Zealand small businesses for 50% of AI advisory costs, capped at NZD 15,000, with an initial allocation of NZD 765,000, applied for through the Regional Business Partner Network and closing 31 January 2027. The maths is simple: a NZD 30,000 advisory and roadmap engagement costs the business 15,000; a NZD 12,000 diagnosis costs 6,000. The subsidy covers advisory fees only, not software subscriptions or running costs. Application details are in our guide to New Zealand government AI funding. 06 ## How BEE Sigma quotes We do not sell standard packages and we do not bill by the hour. The process is a free AIV scan or an AIM implementation assessment first; after seeing real data and interfaces we quote in writing for one testable process, stating scope, approval points, exception handling, acceptance criteria and maintenance. The principle is connect rather than replace: no system is swapped, and the agent lives between the email, WeChat, Lark, Teams, Xero and ERP you already run. To see what these digital employees actually do, read our AI digital employees page. Evidence boundary: prices in the table come from the public pages of AgentWorks.nz (17 May 2026) and EmbedAI (2 September 2026) and from our own market review of 9 September 2026. They are compiled third-party information, not a quote from BEE Sigma or any party; GST is excluded. Durations and ratios for the three cases come from our published case studies; client contract values are not disclosed. ## Primary sources How Much Do AI Agents Cost in NZ? 2026 Pricing Guide (AgentWorks.nz, 17 May 2026) ↗AI implementation cost for NZ SMEs (EmbedAI, 2 Sep 2026) ↗Government backs AI support for small business (beehive.govt.nz) ↗AI companies in New Zealand compared, with price bands (this site, 9 Sep 2026) ↗ FAQ · Quick answers ## How much does an AI solution cost in New Zealand? Public prices fall into four bands: a single-task subscription bot from NZD 2,500 setup plus 150 a month; one custom workflow as a one-off NZD 5,000 to 25,000; a multi-system project connected to Xero or an ERP NZD 25,000 to 90,000; advisory and roadmap NZD 12,000 to 35,000. Running costs after go-live are NZD 100 to 300 a month, GST excluded. MBIE’s AI Advisory Pilot reimburses 50% of advisory fees up to NZD 15,000. Figures are compiled from suppliers’ 2026 public pages and are not a quote. ## How much does an AI digital employee cost per month in New Zealand? Public subscription prices are NZD 2,500 to 12,000 setup plus 150 to 800 a month; custom workflows are usually a one-off NZD 5,000 to 90,000 plus running costs of 100 to 300 a month. GST excluded. ## Is an AI digital employee a one-off purchase or a subscription? Both exist. Boutique studios mostly charge setup plus a monthly fee; implementation teams mostly charge a one-off build plus monthly maintenance. What matters is not the form but whether the quote states process scope, approval points, exception handling and the acceptance method. ## Is there an AI subsidy in New Zealand? Yes. MBIE’s AI Advisory Pilot reimburses eligible small businesses for 50% of AI advisory costs, capped at NZD 15,000, through the Regional Business Partner Network, closing 31 January 2027. It covers advisory fees only, not software subscriptions. ## Why do some quotes say 2,500 and others 90,000? The 2,500 entry point is a single-task subscription bot, such as automatic enquiry replies; the 90,000 end is a multi-process project connected to Xero, ERP and approval systems. The gap is discovery, integration, change management and maintenance; the model itself is a small share. RS BEE Sigma Research Mapping the AI implementation market, providers and policy across New Zealand and China — grounded in checkable facts. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 技术实施机构怎么选｜BEE Sigma 业务流程落地 | BEE Sigma URL: https://beesigma.com/ai-implementation-partner-nz Summary: 选择新西兰 AI 技术实施机构：比较研究、咨询与实施团队的职责，核对系统接入、试点验收、数据权限与上线维护。BEE Sigma 为新西兰与中国企业提供中英双语 AI 工作流实施，保留现有系统与人工审批。 新西兰 AI 技术实施机构怎么选｜BEE Sigma 业务流程落地 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 NEW ZEALAND & CHINA · AI IMPLEMENTATION ## 新西兰 AI 技术实施机构：从选择团队到业务流程验收 BEE Sigma（BEEBEE AI 企业服务品牌）是服务新西兰与中国企业的落地陪跑型团队。我们把 AI Agent 接入现有邮件、Teams、Slack、微信、飞书与 ERP，实施获客、客户跟进、财税与 Golden Flow 定制工作流。这里说明我们适合承担什么，以及你应该如何核验实施机构的交付能力。 品牌：BEE Sigma · 企业：BEEBEE AI Limited奥克兰 · 新西兰与中国中文与英文交付 免费 AIV 诊断AIM 落地评估 DELIVERY · 实施边界 ## 选择实施团队，先看交付证据 01 ## 先区分研究、咨询与实施 研究机构侧重方法探索与技术验证；咨询团队侧重路线图和业务判断；实施团队需要把接口、权限、异常处理和人员操作连成实际流程。若目标是接入现有系统并投入日常使用，应要求候选团队展示从输入到人工确认再到结果记录的完整流程。 02 ## 明确一个可验收的试点 先选一条重复发生、规则较清楚、数据可获得的流程。交付前写明流程范围、输入样本、人工审批点、异常处理、完成时效和验收方法。周期与费用需在看过实际数据和接口后确定，不以演示效果代替生产验收。 03 ## 核对权限、日志与退出方式 确认谁可以读取或写入数据，哪些动作必须审批，失败后由谁处理。把数据导出、访问权限撤销、运行文档和维护责任写进交付范围；涉及付款、发票、合同和申报的操作保留人工决定。 04 ## 查看与业务相近的案例 本站案例包含新西兰教育招生、建筑发票处理与应收账款、房贷客户跟进，以及中新跨境贸易与物流。案例已脱敏，用于说明流程范围；具体客户参考、测量数据及可披露材料需按授权与项目情况确认。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） 了解我们的新西兰 AI 落地实践案例，再对照企业流程自动化解决方案选择试点。需要先确定方向，可读新西兰 AI 咨询服务；需要分阶段扩展，可看企业智能化转型。BEE Sigma 的原则是连接而不是替换、实干而不是论道、陪跑而不是撤场。 FAQ · 选择与核验 ## 新西兰企业选择 AI 技术实施机构的常见问题 ## 新西兰 AI 技术实施机构怎么选？ 先区分你的需求是研究验证、咨询规划还是实际系统实施。若要投入日常业务，要求团队提交流程范围、接口接入方案、人工审批点、异常处理、测试样本和维护责任，并用自己的业务样本验收。BEE Sigma 提供现有系统内的 AI 工作流实施，服务新西兰与中国企业。 ## BEE Sigma 是研究机构还是商业实施团队？ BEE Sigma 是 BEEBEE AI 的企业服务品牌，背后是商业 AI 落地陪跑型团队，不是大学或政府研究机构。服务覆盖 GTM 获客、CRM 跟进、财税流程和 Golden Flow 定制工作流，由中英双语团队交付。 ## AI 技术实施和 AI 咨询有什么区别？ 咨询解决先做什么、为什么做和如何界定范围；实施把流程接入真实系统，完成权限配置、测试、异常处理、人员交接和运行维护。可以先评估再实施，双方应在合同中写明各阶段产物与验收标准。 ## 可以先看新西兰 AI 落地实践案例吗？ 可以。本站独立案例页展示教育、建筑、房贷服务与跨境业务的脱敏流程。公开案例是 BEE Sigma 的交付说明，不代表独立审计；是否能提供客户参考或进一步材料取决于客户授权。 ## 实施 AI 必须换掉现有 ERP 或 CRM 吗？ 不一定。BEE Sigma 优先评估现有系统的接口和权限，通过 Agent 连接邮件、协作工具及业务系统。无法安全接入的环节应先说明限制，不应在评估前承诺所有系统都能自动写入。 ## 怎样联系 BEE Sigma，并开始实施评估？ 可从本站链接进入免费 AIM 落地评估，新西兰与澳洲业务可邮件联系 brinny@beebee.ai，中国大陆、香港与台湾业务可联系 anna@2brain.ai。奥克兰办公地址为 B:Hive, 74 Taharoto Road, Smales Farm, Takapuna。团队服务新西兰与中国，具体交付范围、报价和维护安排在评估后书面确认。 ## 把第一条业务流，交给 Agent 试试 先通过免费 AIM 评估确定适合试点的业务流程；范围、报价与验收标准在实施前书面确认。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI Implementation New Zealand: How to Choose an AI Implementation Partner | BEE Sigma URL: https://beesigma.com/ai-implementation-partner-nz-en Summary: Choose an AI implementation partner in New Zealand: assess integration, pilot acceptance, access controls and support. AI Implementation New Zealand: How to Choose an AI Implementation Partner | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan NEW ZEALAND & CHINA · AI IMPLEMENTATION ## AI implementation in New Zealand: choosing a partner that ships BEE Sigma (BEEBEE AI’s enterprise services brand) is an AI implementation partner in Auckland, New Zealand, serving businesses across New Zealand and China. We build AI agents that connect to existing email, Xero, ApprovalMax, Teams, WeChat and Lark systems, quote a fixed scope in writing after a free AIV scan, and stay involved after go-live. We connect agents to existing email, Teams, Slack, WeChat, Lark and ERP systems for acquisition, CRM, finance and custom Golden Flow workflows. Use this page to assess our scope and the evidence you should request from an implementation partner. Brand: BEE Sigma · Company: BEEBEE AI LimitedAuckland · New Zealand and ChinaEnglish and Chinese delivery Free AIV ScanAIM Assessment DELIVERY · SCOPE ## Choose a team by its delivery evidence 01 ## Separate research, advice and implementation Research explores methods and technical feasibility. Advisory teams help define priorities and business decisions. Implementation connects interfaces, permissions, exception handling and staff actions into a working process. Ask to see the full path from input through human approval to a recorded result. 02 ## Define one testable pilot Start with a repeated process, clear rules and available data. Agree its scope, input samples, approval points, exception handling, turnaround expectations and acceptance method before delivery. Timing and cost depend on real data and interfaces; a demo is not production acceptance. 03 ## Check controls and handover Specify who can read or write data, which actions need approval and who handles failures. Include data export, access revocation, operating documentation and support ownership in the delivery scope. Payments, invoices, contracts and filings retain human decisions. 04 ## Request relevant case evidence Our published cases cover New Zealand education enquiries, construction invoices and receivables, mortgage-client follow-up, plus cross-border trade and logistics. Cases are anonymised to explain process scope. Client references, measured outcomes and further materials depend on permission and project availability. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) Explore our AI implementation case studies and workflow automation solutions. For initial direction, see AI consulting in New Zealand; for staged expansion, see enterprise AI transformation. BEE Sigma connects existing systems, implements working processes and stays involved after launch. FAQ · SELECTION ## Questions to ask an AI implementation partner ## How do I choose an AI implementation partner in New Zealand? First decide whether you need research, advice or operational implementation. For live operations, request a scope, integration plan, approval points, exception handling, test samples and support ownership. Test with your own business inputs. BEE Sigma implements AI workflows in existing systems for New Zealand and China businesses. ## Is BEE Sigma a research institute or a commercial implementation team? BEE Sigma is the brand of BEEBEE AI Limited and a commercial hands-on implementation team, not a university or government research institute. Its bilingual services cover GTM acquisition, CRM follow-up, finance and custom Golden Flow workflows. ## What is the difference between AI consulting and implementation? Consulting determines priorities and scope. Implementation connects real systems and covers permissions, testing, exceptions, staff handover and operational support. Agree the deliverables and acceptance criteria for each stage in writing. ## Can I review New Zealand AI implementation case studies? Yes. Our dedicated case page describes anonymised education, construction, mortgage services and cross-border workflows. These are BEE Sigma delivery descriptions, not independent audits. Further evidence and client references depend on customer permission. ## Do we need to replace our ERP or CRM? Not necessarily. BEE Sigma first assesses the interfaces and permissions of existing systems. Agents can connect collaboration tools and business systems where supported. Integration limits should be documented before anyone promises automated write access. ## How do I contact BEE Sigma for an implementation assessment? Use the free AIM assessment linked on this site or email brinny@beebee.ai for New Zealand and Australia, or anna@2brain.ai for mainland China, Hong Kong and Taiwan. Our Auckland office is at B:Hive, 74 Taharoto Road, Smales Farm, Takapuna. We serve New Zealand and China; delivery scope, price and support are agreed after assessment. ## Hand your first workflow to the agents Start with a free AIM assessment to identify a suitable pilot. Agree scope, price and acceptance criteria before implementation. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # BEE Sigma · 可管理的企业 Agent 平台｜新西兰与中国 AI 落地与咨询 URL: https://beesigma.com/ Summary: 引入 AI，无需更换任何系统。BEE Sigma 是 BEEBEE AI 旗下服务新西兰与中国两地企业的 AI 科技咨询与落地品牌（团队在奥克兰），把 12 个职能 Agent 接入企业已有的微信、飞书、Slack、Teams、邮件与 ERP，从获客、签约到回款，中英双语交付。 BEE Sigma · 可管理的企业 Agent 平台｜新西兰与中国 AI 落地与咨询 BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 AI NATIVE · 可管理的企业 AGENT 平台 ## 引入 AI，无需更换任何系统。 BEE Sigma 将一支分工明确的 Agent 团队，接入企业已有的微信、飞书、Slack、Teams、邮件与 ERP——系统不换，流程不动，团队只负责提出要求、做出决策。 获取免费 AIV 诊断 观看 3 分钟演示 12个职能分工的 Agent 3+1条工作流 · 三条标准 + Golden Flow 2个市场 · 新西兰与中国 8+个行业交付场景 Command · 你只需要说一句话 ## 团队负责提出要求，Agent 负责跑完流程 你说 WECHAT · LARK · SLACK · TEAMS · ERP● 12 AGENTS ONLINE GTM获客与培育—线索→ CRM转化与签约—合同→ 财税回款与合规—现金 ✓ +GOLDEN FLOW你的产金流 · 定制 01 · GTM 工作流 ## 获客与培育重新定义市场触达的流水线 G1 AIV 执行官G2 市场雷达 G3 内容官G4 猎手 产出 · 合格线索 G1 ## AIV 执行官 诊断并优化官网的 SEO 与 GEO，让搜索引擎和 AI 都推荐你，而不是竞品。产出：AI 可见度评分 + 修复清单 G2 ## 市场雷达 持续盯行业、竞品与政策，配置只需一张飞书卡片，每天早上推简报进群。产出：每日情报 + 行动建议 G3 ## 内容官 拿雷达的情报出选题，按品牌声音写公众号、按平台习惯改小红书。产出：多平台内容 · 发布前人审 G4 ## 猎手 富化线索、按 ICP 打分，没达标的继续培育，达标的移交 CRM 情报官。产出：打好分的合格线索 02 · CRM 工作流 ## 转化与签约让每一单，推进得有记忆 C1 情报官C2 方案师 C3 跟单官C4 交接官 产出 · 干净的合同交接 C1 ## 情报官 接住线索就开始摸底：公开信息、历史交集、决策链，一页看懂。产出：客户档案 · 含历史记忆 C2 ## 方案师 调同类客户案例，按这个销售自己的措辞习惯，生成能签的提案。产出：提案初稿 · 风格对齐 C3 ## 跟单官 把每轮沟通归档成时间线，关键承诺推给销售一键确认，每轮给策略。产出：时间线 + 下一步建议 C4 ## 交接官 签约后汇编全部承诺与特殊条款，写回订单，交付团队不用再追问。产出：交付简报 · 写回经审批 03 · 财税工作流 ## 回款与合规把现金流，变成确定性 F1 单据官F2 对账官 F3 催款官F4 税务官 产出 · 回款与合规留痕 F1 ## 单据官 票据识别、查重校验，签约即备好开票资料——发票开出前必经人手。产出：开票资料 · 进项核验 F2 ## 对账官 银行流水对 ERP 自动核销，对不上的列成差异清单给会计处理。产出：核销结果 + 差异清单 F3 ## 催款官 盯账期，复用 CRM 侧的关系记忆分级措辞——长期客户温和提醒，频繁逾期者措辞升级。产出：催款草稿 · 发送前人审 F4 ## 税务官 申报日历、凭证归集、预填税表，只做准备和提醒——申报永远由人执行。产出：申报准备包 · 只备不报 04 · GOLDEN FLOW 定制工作流 ## 你的产金流真正的金子，在你独特的业务模式里 ⊕ 独立于三条标准流 来源 · 你的业务模式 产出 · 按业务结果验收 GF ## 什么是 Golden Flow GTM、CRM、财税是每家企业都有的通用流；真正让你赚钱的，是你独有的那条——报价的算法、排产的经验、老师傅的判断、行业里攒下的资源。把它交给一位 AI 数字员工。产出：只属于你的 Agent 工作流 01 ## 它长什么样 案例里的工地工时记录、微信群订单到顺丰面单、H1 合规计算，都不在三条标准流里——它们就是各家的 Golden Flow。参考：脱敏案例与客户反馈 02 ## 怎么找到它 从免费 AIM 评估开始：把业务流摊开，找出最值得交给 Agent 的那一条，先小范围试点。产出：切入点 + 验收口径 03 ## 怎么交付 复用平台的记忆、审批、日志与 IM 接入，只为你的流程新建缺的 Agent——高风险动作依然人来拍板。产出：定制 Agent 编队 · 按结果验收 人工确认点贯穿始终：官网改动、内容发布、报价折扣、写回订单、发票开出、一切申报——高风险动作，Agent 提请，人来拍板。 Casebook · 脱敏案例 ## 九个服务项目，按交付状态如实标注 来自新西兰与中国两地的服务项目。应客户要求隐去企业名称；每张卡片标注交付状态（已上线、部分上线、试点 · 演示、开发中），数字取自标注日期的项目记录。 产品案例 · 老板分身已上线（产品） ## 老板分身：把老板的判断装进团队的 AI 智能体 BossClone · 新西兰与中国 6 家客户的使用记录 15–30 分钟六层访谈生成 SOUL.md，3 步装进飞书或企业微信里的智能体；同事在群里先按老板的标准拿到判断，越界的事先问老板本人。看完整案例：工作原理、实操截图与客户反馈 → 教育与培训NZ试点 · 演示 ## 新线索进来，先有一封回信和预约链接 新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） Lead 进入 CRM→自动首封邮件 + 预约链接→预约者进顾问队列 ✕ 原来两千多条新线索堆在 CRM 里（客户在 2026-07-31 会议上的估算），顾问只能逐个打电话，分不出优先级。 试点：新线索进入 Zoho 后自动发个性化邮件与预约链接，用“是否预约”筛出高意向交给顾问 Agent 已接入 Microsoft Teams；录取后未推进签证的学生由 Agent 标记并起草跟进邮件（测试环境验证） 查看完整案例说明 → 教育与培训CN试点 · 演示 ## 一个学生的 18 份散落资料，变成一张可检索的成长档案 中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） 原始资料→Agent 逐份摘要→多维表格索引→教练按阶段调用 ✕ 原来测评报告、教练反馈、三年规划散在 PDF 和 Word 里，角色之间交接有断点。 首个学生样板：18 份原始资料加 10 份团队沉淀，建成 28 条结构化索引 服务方案与专业判断由机构提出并确认，AI 只负责结构化与表达 查看完整案例说明 → 建筑与工程NZ已上线 ## 应付发票，不再靠人追着跑 新西兰 · 建筑总承包商 · 应付发票自动分流 发票识别→匹配 PO 与审批人→提交 ApprovalMax→状态回写仪表盘 ✕ 原来发票人工录入，审批卡在谁那里没人知道，逾期了才被发现。 Agent 识别字段、归档、匹配采购单并提交审批；审批与付款仍由人在 ApprovalMax 与 Xero 里完成 一个会计月 798 份文档中 730 份自动分流（91.48%，2026-09-01 项目复盘口径）；Admin 与审批人两版仪表盘 完整实录：发票自动分流 91% → 建筑与工程NZ部分上线 ## 开票、催收、风控：Agent 起草，人最后一按 新西兰 · 建筑公司 · 应收账款自动化 报价单→Xero 开票并发送→每日拉取逾期→催款草稿三档→风险分级卡片 ✕ 原来开票靠人工录入，逾期账款靠财务翻 Xero 账龄表，催款邮件语气全凭当时心情。 Agent 读取 Google Drive 报价单，在 Xero 建发票并发送（2026-07 MVP 上线） 每天 09:00 拉取 Xero 逾期数据，生成三档语气的催款草稿，只生成不自动发送；四级风控卡片供管理层决策（2026-09 演示阶段） 查看完整案例说明 → 建筑与工程NZ已上线 ## 工头一句话，工时进系统 新西兰 · 南岛建筑施工公司 · 工地工时与人工成本 飞书群一句话→Agent 解析写表→工头确认→按时薪自动计算→成本看板 ✕ 原来工时靠工头下班后回忆、再汇总进 Excel，老板看不到每个项目的人工成本。 工头在已有的飞书群里说一句话，Agent 写进日工时明细与工地日志；时薪计算走多维表格自动化，不经过大模型（2026-04-08 启动，06-15 全公司使用） 2026-06-27：20 个项目、334 条日常记录、累计 1 万多小时工时进入系统；工资与成本按角色分权限查看 完整实录：工地工时从 Excel 到飞书群 → 跨境贸易与物流NZ ⇄ CN已上线 ## 微信群里的订单，第一次有了闭环 中新跨境 · 保健品零售 · 微信群订单到顺丰面单 群内订单→拆单与核验卡→人核对四项说“审核通过”→顺丰建单与证件归档→面单回调回写 ✕ 原来订单埋在群聊里，业务员要把同一笔订单抄进总表、顺丰模版和证件归档，再把面单一张张发回。 Agent 解析微信群订单、映射 SKU、拆单并生成核验卡；人核对收件、商品、付款、证件四项后下达“审核通过”，Agent 才调用顺丰接口建单（2026-09-02 生产首单） 顺丰出库后主动推送面单，Agent 归档并回写多维表格（2026-09-09 打通）；ERP 与微信均未更换 完整实录：微信群下单到顺丰面单 → 跨境贸易与物流NZ ⇄ CN部分上线 ## 从建档到托书，一条货代售前流不再靠人盯 中新航线 · 国际货代 · 售前到目的港三条工作流（分阶段交付） CRM 建档→订单创建→报价 PDF 人工审批→确认邮件→托书生成 ✕ 原来订单创建、报价审批、托书交接没有统一规则，前期反复沟通产生大量无效订单。 售前流：建档、建单、报价审批、确认邮件、托书生成已串联，报价由 sales support 审批后才建单（2026-09 交付） 始发港 Pre-Alert 文件校验与目的港 ETA、放行提醒开发中，2026-09-15 阶段评审 查看完整案例说明 → 金融与商业服务NZ部分上线 ## 从一封生日祝福邮件开始的老客户触达 新西兰 · 房贷顾问团队 · 内容运营与客户触达 企业邮箱授权→生日自动祝贺→小红书选题与出图→每日房贷市场简报 ✕ 原来老客户生日靠顾问自己想起来；小红书账号内容积累少，提炼不出风格。 客户生日祝贺邮件经 Outlook 授权后按日自动发出（2026-08 打通） 小红书选题池与出图规则建立，顾问与 Agent 单聊即可产出内容；每天 08:00 推送房贷、利率与 refix 政策简报 查看完整案例说明 → 金融与商业服务CN试点 · 演示 ## 销售一句话，AI 填好 CRM 预览，人确认再落库 中国 · 园区与场馆招商 · 对话式 CRM 录入（MVP 演示阶段） 飞书群一句话→七项字段预览→回复“确认并写入”→客户 / 项目 / 跟进三表 ✕ 原来跟进记在聊天和脑子里，CRM 字段常年空着，历史数据散在各销售的 Excel。 销售在飞书群描述拜访情况，Agent 解析、查重并生成七项字段预览，精确回复“确认并写入”才落库（MVP v3，2026-09 演示） 联系人五类角色与关系分级，一键生成决策链图（2026-09-10 测试通过） 查看完整案例说明 → 查看全部 9 个案例 本页案例均来自 BEE Sigma 在新西兰与中国的真实服务项目，应客户要求隐去企业名称。为准确起见，每张卡片注明交付状态：已上线指工作流已在客户日常业务中运行并由客户团队使用；部分上线指部分环节已在用、其余在演示或开发中；试点 · 演示指已在测试环境或演示会上跑通、尚未全量投入日常使用；开发中指处于方案或开发阶段。所列数字均为记录中的项目复盘口径并标注日期，部分未经客户书面确认；项目效果因客户现有系统、数据与配合程度而异，不构成对其他企业的承诺。页面内容截至所标日期，之后系统由客户自行运营，我们不持有最新数据。 Proof · 眼见为实 ## 两支互动演示，完整走一遍从线索到回款 不罗列功能清单，只展示真实流程：同一个客户，从市场雷达中出现，到财务开出发票。 GTM 篇 · 飞书 EP.01 · 02'45" ## 《一条获客流水线》 从 AI 可见度诊断 32 分到 71 分：市场雷达捕捉竞品调价窗口，内容生成、线索评分与移交全程在飞书完成——配置只是一张卡片。 进入互动演示 CRM 篇 · 微信 EP.02 · 03'00" ## 《恒达机械这一单》 从一条摸底指令开始，14 轮沟通、一次关键确认，直至交付简报与发票草稿——销售全程未打开 CRM，财务无需催要任何资料。 进入互动演示 Trust · 深受企业信赖 ## BEE Sigma深受两地企业信赖 新西兰与中国两个市场、八个以上行业交付场景。客户与合作伙伴既有本地服务企业，也有跨境贸易与金融团队。 NZSEG 3EYES Construction Coconut Loans JL Roofing 太平人寿 太平洋保险 普元信息 Novo Education AUNZ International CHINZ 秦远物流 Mars Living Briv Construction 财迷游言 华樱教育 琪石培训 蜗牛口腔 译讯科技 MedMed PawPrint Studio 浙江智汇通 北京万馆 九洲昌隆 泓瑀泰达 Delivery notes · 来自项目记录 ## 交付记录里的几个片段 以下摘要取自项目记录与复盘纪要，日期已标注，企业名已脱敏；不是客户评价，也不构成客户背书。 ## 一个会计月 798 份发票文档，730 份自动分流 自动分流比例 91.48%，人工复核降到 68 份；审批与付款仍由人在 ApprovalMax 与 Xero 里完成。2026-09-01 项目复盘口径，未经客户书面确认。 新西兰 · 建筑总承包商应付发票 · 已上线 ## 10 周，工地工时从 Excel 搬进飞书群 2026-04-08 启动，06-15 全公司使用；06-27 汇报时 20 个项目、334 条日常记录、1 万多小时工时在多维表格里。之后系统由客户自行运营。 新西兰 · 南岛建筑公司工地工时 · 已上线 ## 90 天，微信群订单到顺丰生产首单 2026-06-04 启动，09-02 生产首单，09-09 顺丰面单回调打通；人核对四项后说“审核通过”，Agent 才建单。截至 09-11 日常工单在正常处理。 中新跨境 · 保健品零售订单到面单 · 已上线 ## 11 天，一条销售对账工作流从立项到上线 2026-07-20 立项，07-31 上线；两天培训场景下由智能体产出 30 多份客户分析报告（2026-07-27 季度复盘纪要）。 中国 · 金融培训机构销售对账 · 已上线 Founder · 创始人 ## 周品 Pin Zhou BEEBEE AI / 2BRAIN 创始人 25 年互联网行业经验，历经 3 次创业，曾任两家美股上市公司（百度、猎豹移动）高管。 长期专注 AI 落地、企业咨询陪跑与组织能力建设——不是做演示，而是把流程真正跑起来。 AI 咨询与培训累计服务 60+ 企业与机构、赋能 1000+ 职场人士，覆盖教育机构、华人企业主与成长型团队。 关注的不是“讲 AI”，而是“把 AI 变成真正能跑起来的业务系统”。 Governance · 可管理 ## "可管理"不是口号，是写进流程的边界 这正是企业敢把业务交给 Agent 的前提。负责任 AI 承诺 → ⬡ ## 谁能做什么 Agent 权限逐条授权，逐条可收回。 ⬡ ## 谁来审批 高风险动作必经指定同事确认。 ⬡ ## 出错怎么办 每个动作留痕、可追溯、可回滚。 ⬡ ## 数据在哪里 本地化部署，数据不出你的企业。 Start · 怎么开始 ## 三步落地，每一步都有验收和回退 不推倒现有系统，不要求全员培训；先运行一条业务线，验证见效后再扩展。 01 ## 免费 AIV 诊断 评估企业官网在 AI 检索中的可见度：谁在被推荐、为何未被收录。一份报告，同时厘清最值得优先改造的业务环节。 02 ## 单条工作流试点 选择价值最高的一条（多数客户从 CRM 跟单或财务对账切入），在最小范围内跑通闭环：验证可用性，明确确认人与退出机制。 03 ## 接入真实系统 验证有效后接入企业微信、飞书、Teams、Slack 与 ERP，本地化部署上线；持续跟踪时效、复用率与错误率，逐条扩展。 FAQ · 常见问题 ## 决定之前，你可能想问的八个问题 ## BEE Sigma 是什么？ BEE Sigma 是 BEEBEE AI 的企业服务品牌，背后是一支 AI 科技咨询与工程落地团队，也是一个可管理的企业 Agent 平台（我们交付的 Agent 也常被叫作 AI 数字员工）：我们把一支分工明确的 Agent 团队（12 个职能 Agent、三条标准工作流 + Golden Flow 定制工作流）接入企业已有的微信、飞书、Slack、Teams、邮件与 ERP，从获客、签约到回款形成可复用、可验收的 AI 工作流和知识系统——不是又一个要学的新软件。 ## BEE Sigma 适合哪些企业？ 适合客户跟进、知识管理与交付高度依赖人工经验的企业：教育与培训、建筑与工程、跨境贸易与物流、金融与商业服务等都有已交付案例。一个简单的判断标准：如果你的团队还在群聊里复制、往表格里粘贴、靠个人记忆跟客户，就适合。如果还在纠结要不要先换一套管理软件，可以先读新西兰企业管理软件比较，多数企业不必换系统。 ## 和普通 AI 培训有什么不同？ 普通培训的重点是教工具，学完回到工位一切照旧。BEE Sigma 对业务结果负责：从一个具体场景切入，陪你完成流程诊断、知识沉淀、Agent 工作流搭建和上线验证——高风险动作始终由人审批，每一步留痕可回退。 ## 一次 AI 落地项目通常交付什么？ 通常包括：业务流程诊断、知识资产整理、Agent 工作流设计与配置（从 GTM 获客、CRM 跟单或财税对账中选价值最高的一条切入）、与现有系统的接入、团队培训，以及一套明确的验收指标。支持本地化部署，数据不出企业。 ## 项目一般多久能看到效果？ 我们建议先用 2–4 周跑通一个最小业务闭环——多数客户从 CRM 跟单或财务对账切入，见效最快；复杂场景通常用 3–6 个月逐条工作流扩展。每一步都有验收和回退，不满意可以停。 ## 客户需要配合什么？ 提供真实业务流程、历史案例、常见问题、交付材料，并安排关键人员访谈——我们负责把这些隐性经验转成 Agent 可用的知识。前期每周大约需要业务负责人 2–3 小时。 ## 在教育行业能解决什么问题？ 帮教育机构把招生咨询建成不断线的漏斗、给每位顾问配一个成交辅助（会前匹配、会中提示、会后画像）、沉淀资深顾问的经验训练新人、规范家校沟通，并把教学与咨询 SOP 变成可复用的知识资产——新西兰与中国都有已交付的教育案例。 ## BEE Sigma 的服务地区在哪里？ 服务中国、新西兰、澳大利亚、泰国及东南亚市场，支持中英文业务场景。奥克兰办公室位于 B:Hive, Smales Farm（Takapuna），中国团队覆盖成都与北京。 ## BEE Sigma 是哪家公司的品牌？正规吗？ BEE Sigma 是 BEEBEE AI 的企业服务品牌，运营主体 BEEBEE AI Limited 在新西兰公司注册处登记，NZBN 9429051781781，办公地址 B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland 0622，Google 商家档案可查。BEE Sigma 不是加密货币、投资平台或交易软件，与任何同名代币无关；我们只做企业 AI 咨询与工作流落地，已发表的案例均为标注日期的脱敏项目记录。 ## 新西兰 AI 技术应用有哪些真实场景？ 按 BEE Sigma 2026 年标注日期的项目记录，新西兰 AI 技术应用集中在四类流程：供应商发票分流（建筑总承包商，一个会计月 798 份文档中 730 份自动分流，2026-09-01 复盘口径）、工地工时录入（南岛建筑公司，10 周全公司使用）、跨境订单到物流面单（保健品零售，90 天生产首单）、销售对账（金融培训机构，11 天上线）。全部不换系统，Agent 接在 Xero、ApprovalMax、飞书、微信与 ERP 之间；付款、审批与对外承诺留给人。完整记录见AI 落地实践案例与企业如何利用 AI 提升效率。 ## BEE Sigma，企业拥抱 Agent 时代的方式 两个免费入口，各花几分钟——先看清自己，再决定从哪里开始。 AIV · AI VISIBILITY ## 免费 AIV 诊断 看清企业官网在 AI 检索中的可见度：谁在被推荐、为何未被收录，报告同时指出最值得优先改造的业务环节。 开始诊断 AIM · 落地评估 ## 免费 AIM 评估 回答几个关于业务现状的问题，快速看清你的企业落地 AI 最合适的切入角度——从哪条业务流开始，先解决什么。 开始评估 企业微信 · 扫码直接沟通 联系 · 新西兰与澳大利亚brinny@beebee.ai 联系 · 中国大陆、香港与台湾anna@2brain.ai 本地化部署 · 数据不出企业 · 关键动作人审批 · 新西兰注册运营 BEESigma AI Native · 可管理的企业 Agent 平台。让 Agent 接入企业现有系统，团队专注于提出要求与做出决策。 办公地址：B:Hive, 74 Taharoto Road,Smales Farm, Takapuna, Auckland, New Zealand联系：发邮件给我们 ## 产品 GTM 工作流 ↗CRM 工作流 ↗财税工作流 ↗Golden Flow 定制 ↗AIV 诊断 ↗AIM 落地评估 ↗ ## 看看 博客 ↗脱敏案例 ↗GTM 演示分镜 ↗CRM 演示分镜 ↗AI 咨询服务 ↗AI 实施服务 ↗ ## 公司 关于创始人 ↗常见问题 ↗负责任 AI ↗合作流程 ↗新西兰/澳洲 · brinny@beebee.ai ↗中国/港台 · anna@2brain.ai ↗ © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · Google 商家档案 English ===== # BEE Sigma · AI Consulting & AI Implementation in New Zealand | Enterprise Agent Platform, NZ & China URL: https://beesigma.com/en Summary: BEE Sigma, BEEBEE AI’s enterprise services brand, offers AI consulting and implementation in Auckland for NZ and China, with agents plugged into your systems. BEE Sigma · AI Consulting & AI Implementation in New Zealand | Enterprise Agent Platform, NZ & China BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan AI NATIVE · THE MANAGEABLE ENTERPRISE AGENT PLATFORM ## Adopt AI.Replace nothing. BEE Sigma plugs a team of specialized agents into the systems you already run — Teams, Slack, WeChat, Lark, email and your ERP. No migration, no process change. Your team just asks, and decides. Get a Free AIV Scan Watch the 3-min Demo 12specialized agents 3+1workflows · 3 standard + Golden Flow 2markets · NZ & China 8+industry deployments Command · One sentence is all it takes ## Your team asks. The agents run the process. You WECHAT · LARK · SLACK · TEAMS · ERP● 12 AGENTS ONLINE GTMAttract & nurture—leads→ CRMConvert & close—contract→ FINCollect & comply—cash ✓ +GOLDEN FLOWyour money-making flow 01 · GTM WORKFLOW ## Attract & NurtureAn assembly line for market reach G1 AIV OfficerG2 Market Radar G3 Content AgentG4 Hunter Output · qualified leads G1 ## AIV Officer Audits and optimizes your site’s SEO and GEO so search engines and AI recommend you — not your competitors.Output: AI visibility score + fix list G2 ## Market Radar Watches your industry, competitors and policy around the clock. Setup is one Lark card; a briefing lands in your group every morning.Output: daily intel + suggested actions G3 ## Content Agent Turns radar intel into topics, writes in your brand voice, adapts for each platform’s conventions.Output: multi-platform content · human-approved G4 ## Hunter Enriches leads and scores them against your ICP. Below-bar leads keep nurturing; qualified ones hand over to CRM Intel.Output: scored, qualified leads 02 · CRM WORKFLOW ## Convert & CloseEvery deal moves forward with memory C1 Intel AgentC2 Proposal Agent C3 Deal AgentC4 Handover Agent Output · clean contract handover C1 ## Intel Agent Starts digging the moment a lead arrives: public records, past interactions, decision chain — one page tells the story.Output: client dossier · with full history C2 ## Proposal Agent Pulls comparable client cases and drafts a signable proposal in the salesperson’s own voice.Output: proposal draft · voice-matched C3 ## Deal Agent Files every exchange into a timeline, pushes key commitments for one-tap confirmation, suggests strategy each round.Output: timeline + next-step advice C4 ## Handover Agent After signing, compiles every promise and special term and writes it back to the order. Delivery never has to ask twice.Output: delivery brief · approved write-back 03 · FINANCE WORKFLOW ## Collect & ComplyTurn cash flow into certainty F1 Invoice AgentF2 Reconciliation F3 CollectionsF4 Tax Agent Output · cash in, audit trail kept F1 ## Invoice Agent Reads documents, checks duplicates, preps invoicing the moment a contract signs — no invoice leaves without human sign-off.Output: invoice pack · input verification F2 ## Reconciliation Agent Matches bank statements against ERP automatically; anything unmatched becomes a discrepancy list for your accountant.Output: matched entries + discrepancy list F3 ## Collections Agent Watches payment terms and reuses CRM relationship memory to calibrate tone — gentle with loyal clients, firmer with repeat late-payers.Output: dunning drafts · human-reviewed F4 ## Tax Agent Filing calendar, document collection, pre-filled forms. Preparation and reminders only — filing is always done by a human.Output: filing prep pack · prep only 04 · GOLDEN FLOW · CUSTOM BUILD ## Your Golden FlowThe real gold is in your unique business model ⊕ Beyond the three standard flows Source · your business model Output · verified by business results GF ## What is a Golden Flow GTM, CRM and Finance are flows every business shares. What actually makes you money is the flow only you have — your quoting logic, scheduling experience, veteran judgment, the resources your years in the trade have earned. Hand it to an AI digital employee.Output: an agent workflow that exists nowhere else 01 ## What it looks like The training-and-assessment loop, interview practice and knowledge Q&A in our client voices sit outside the three standard flows — each one is that client’s Golden Flow.See: casebook & client voices above 02 ## How we find yours Start with the free AIM assessment: lay your business flows out, pick the one most worth handing to agents, and pilot it small.Output: entry point + acceptance criteria 03 ## How we deliver We reuse the platform’s memory, approvals, logging and IM integrations, and only build the agents your process is missing — humans still sign off on high-risk actions.Output: a custom agent squad · verified by results Human checkpoints throughout: site changes, content publishing, discounts, order write-backs, invoicing, every filing — for high-stakes actions, agents propose, people decide. Casebook · Anonymized Cases ## Nine engagements, each labelled with its delivery status Engagements from New Zealand and China. Company names withheld at clients’ request; each card shows its delivery status (live, partly live, pilot/demo, in development) and figures come from dated project records. PRODUCT CASE · BOSS CLONELive (product) ## Boss clone: the owner’s judgement inside the team’s AI agent BossClone · records from 6 clients in New Zealand and China A 15–30 minute six-layer interview generates SOUL.md, installed in a Lark or WeCom agent in 3 steps. Colleagues get a first judgement by the owner’s standards in their group chat, and anything out of scope goes to the owner.Read the full case: how it works, walkthrough and client feedback → Education & TrainingNZPilot · Demo ## A first reply and a booking link within minutes of a new lead NZ · International education group · lead screening (phase-one pilot) Lead enters CRM→Auto first email + booking link→Booked leads to counsellor queue ✕ BEFOREMore than two thousand new leads sat in the CRM (the client’s estimate at a 2026-07-31 meeting); counsellors could only call one by one with no way to prioritise. Pilot: new Zoho leads get a personalised email and booking link; those who book are routed to a counsellor Agent live in Microsoft Teams; admitted students with stalled visas are flagged with a draft follow-up (validated in a test environment) Read the full case → Education & TrainingCNPilot · Demo ## Eighteen scattered files, one searchable student record CN · Study-abroad consultancy · coaching-service documentation (co-design phase) Source files→Agent summaries→Lark Base index→Coach retrieval by stage ✕ BEFOREAssessment reports, coach notes and three-year plans lived in PDFs and Word files with gaps at every hand-off. First pilot student: 18 source files plus 10 team notes indexed into 28 structured records Programme design and professional judgement stay with the firm; AI structures and drafts Read the full case → Construction & TradesNZLive ## Payables that no longer need chasing NZ · Construction head contractor · automatic invoice routing Invoice OCR→PO & approver match→Submit to ApprovalMax→Status write-back ✕ BEFOREInvoices keyed by hand, approvals stuck with no visibility, overdue only discovered after the fact. Agents read, file, match purchase orders and submit approvals; people still approve in ApprovalMax and pay in Xero 730 of 798 documents routed automatically in one accounting month (91.48%, project review of 1 Sep 2026); admin and approver dashboards Full case study: 91% invoice automation → Construction & TradesNZPartly live ## Invoicing, collections and risk: agents draft, people press send NZ · Construction company · receivables automation Quote→Xero invoice & send→Daily overdue pull→Reminder drafts, 3 tones→Risk tier cards ✕ BEFOREInvoices were keyed by hand, overdue accounts meant reading the Xero ageing report, and reminder tone depended on the day. Agent reads quotes from Google Drive, creates and emails the Xero invoice (MVP live Jul 2026) Daily 09:00 pull of Xero overdue items into reminder drafts in three tones, drafts only, never auto-sent; four-tier risk cards for management (demo stage, Sep 2026) Read the full case → Construction & TradesNZLive ## One sentence from the foreman, hours in the system NZ · South Island builder · site hours and labour cost One sentence in Lark→Agent parses and writes→Foreman confirms→Pay rates applied automatically→Cost dashboard ✕ BEFOREHours were recalled by foremen after work and totalled in Excel; the owner could not see labour cost per project. Foremen post one sentence in the Lark group they already use; the agent writes the daily hours detail and site diary, and pay rates are applied by a Base automation, not a language model (kickoff 2026-04-08, company-wide 06-15) 2026-06-27: 20 projects, 334 daily records and over 10,000 hours in the system; payroll and cost visible by role Full case study: site timesheets from Excel to a chat group → Trade & LogisticsNZ ⇄ CNLive ## WeChat-group orders, closed-loop for the first time NZ ⇄ CN · Health-products retail · WeChat orders to SF Express labels Group order→Split & verification card→Human 4-point check→SF order & ID archive→Label callback ✕ BEFOREOrders sat in group chats; one staff member copied each into a master sheet, the SF template and an ID archive, then sent labels back one by one. Agent parses WeChat orders, maps SKUs, splits and issues a verification card; a person checks recipient, product, payment and ID and says ‘review passed’ before the agent calls SF Express (first production order 2 Sep 2026) SF pushes the label after dispatch; the agent archives it and writes back to Lark Base (live 9 Sep 2026). ERP and WeChat unchanged Full case study: WeChat orders to SF Express labels → Trade & LogisticsNZ ⇄ CNPartly live ## From client file to shipping instruction, a pre-sales line nobody has to babysit NZ ⇄ CN lane · Freight forwarder · three workflows, pre-sales to destination port (phased delivery) CRM record→Order creation→Quote PDF approval→Confirmation email→Shipping instruction ✕ BEFORENo shared rules for order creation, quote approval or shipping-instruction hand-off; repeated back-and-forth produced many dead orders. Pre-sales: client record, order, quote approval, confirmation email and shipping instruction chained; quotes released only after sales-support approval (delivered Sep 2026) Origin-port pre-alert checks and destination-port ETA and release alerts in development; stage review 15 Sep 2026 Read the full case → Finance & Business ServicesNZPartly live ## Client touchpoints that start with a birthday email NZ · Mortgage adviser team · content operations & client touch Mailbox authorised→Birthday emails→Xiaohongshu topics & images→Daily mortgage brief ✕ BEFOREClient birthdays depended on an adviser remembering; the Xiaohongshu account had too little content to find a voice. Birthday emails go out automatically through the authorised Outlook mailbox (live Aug 2026) Topic pool and image rules set up; advisers produce posts by chatting with the agent; a daily 08:00 brief on mortgage rates, OCR and refix news Read the full case → Finance & Business ServicesCNPilot · Demo ## One sentence from sales, a CRM preview from AI, saved only after confirmation CN · Venue & park leasing · conversational CRM capture (MVP demo stage) One sentence in Lark→Seven-field preview→Reply ‘confirm and write’→Client / project / follow-up tables ✕ BEFOREFollow-ups lived in chats and heads; CRM fields sat empty; history was scattered across each salesperson’s Excel. Sales describe a visit in a Lark group; the agent parses, de-duplicates and previews seven fields; nothing is saved until they reply ‘confirm and write’ (MVP v3, demoed Sep 2026) Five contact roles and relationship tiers; one-click decision-chain diagram (tested 10 Sep 2026) Read the full case → View all 9 cases The cases on this page come from BEE Sigma’s actual client engagements in New Zealand and China; company names are withheld at clients’ request. For accuracy, each card states its delivery status: Live means the workflow runs in the client’s day-to-day operations and is used by their team; Partly live means some steps are in use while others are in demo or development; Pilot · Demo means it has run in a test environment or demonstration but is not yet in full daily use; In development means it is at the design or build stage. Figures are taken from dated project review records and some have not been confirmed in writing by the client; results depend on each client’s existing systems, data and participation and are not a promise of outcomes for other businesses. Content is current as at the dates shown; systems are subsequently operated by clients and we do not hold later figures. Proof · Seeing is believing ## Two interactive demos, walking from lead to cash No feature lists — just the real flow: one client, from first appearing on the market radar to finance issuing the invoice. GTM · Lark EP.01 · 02'45" ## The Lead Assembly Line From an AI visibility score of 32 to 71: the radar spots a competitor pricing window; content, lead scoring and handover all happen in Lark — setup is a single card. Enter the demo CRM · WeChat EP.02 · 03'00" ## The Hengda Machinery Deal From one briefing request through 14 rounds of communication and a single key confirmation, to the delivery brief and invoice draft — sales never opened the CRM; finance never chased a document. Enter the demo Trust · Chosen by businesses ## BEE Sigma,trusted in two markets Two markets — New Zealand and China — and deployments across eight-plus industries, from local service firms to cross-border trade and finance teams. NZSEG 3EYES Construction Coconut Loans JL Roofing 太平人寿 太平洋保险 普元信息 Novo Education AUNZ International CHINZ 秦远物流 Mars Living Briv Construction 财迷游言 华樱教育 琪石培训 蜗牛口腔 译讯科技 MedMed PawPrint Studio 浙江智汇通 北京万馆 九洲昌隆 泓瑀泰达 Delivery notes · From project records ## A few entries from the delivery record These summaries are taken from project records and review notes, dated and anonymised. They are not client testimonials and do not constitute client endorsement. ## 730 of 798 invoice documents routed automatically in one accounting month An automatic routing rate of 91.48%, with manual review down to 68 documents; approval and payment still done by people in ApprovalMax and Xero. Project review of 1 Sep 2026, not confirmed in writing by the client. NZ · Construction head contractorPayables · live ## 10 weeks from Excel timesheets to a Lark group Kickoff 8 Apr 2026, company-wide use from 15 Jun; at the 27 Jun handover 20 projects, 334 daily records and 10,000+ hours were in Lark Base. The client has run the system since. NZ · South Island builderSite timesheets · live ## 90 days from WeChat-group orders to the first production SF Express order Kickoff 4 Jun 2026, first production order 2 Sep, SF label callback live 9 Sep; a person checks four items and says ‘review passed’ before the agent creates the order. Daily work orders were running normally as at 11 Sep. NZ ⇄ CN · Health-products retailOrders to labels · live ## 11 days from kickoff to go-live for a sales reconciliation workflow Started 20 Jul 2026, live 31 Jul; in a two-day training scenario the agents produced 30+ client analysis reports (quarterly review notes, 27 Jul 2026). CN · Financial training firmSales reconciliation · live Founder ## Pin Zhou FOUNDER · BEEBEE AI / 2BRAIN 25 years in the internet industry, three ventures founded, former executive at two NASDAQ-listed companies (Baidu, Cheetah Mobile). Focused on AI adoption, hands-on advisory and organizational capability — not demos, but processes that actually run. 60+ companies and institutions served and 1,000+ professionals trained across education, Chinese-founded businesses and growth-stage teams. What matters isn’t talking about AI — it’s turning AI into business systems that actually run. Governance · Manageable ## “Manageable” isn’t a slogan — it’s boundaries written into the process This is why businesses dare to hand real work to agents. Our Responsible AI commitment → ⬡ ## Who can do what Agent permissions granted one by one — and revocable one by one. ⬡ ## Who approves High-risk actions always pass through a designated colleague. ⬡ ## When things go wrong Every action is logged, traceable and reversible. ⬡ ## Where the data lives Deployed on your side — data never leaves your company. Start · How it begins ## Three steps to production, each with acceptance criteria and a way back No rip-and-replace, no company-wide training. Run one line of business first; expand once it proves out. 01 ## Free AIV Scan See how visible your site is to AI search: who gets recommended, why you’re missing. One report — and it also pinpoints the business process most worth transforming first. 02 ## One-workflow pilot Pick the highest-value workflow (most clients start with CRM follow-up or reconciliation) and close the loop at minimal scope — proving usability, naming approvers, defining the exit. 03 ## Connect real systems Once proven, connect Teams, Slack, WeChat Work, Lark and your ERP, deploy on your infrastructure, and track turnaround, reuse and error rates as you expand workflow by workflow. FAQ ## Eight questions worth asking before you decide ## What is BEE Sigma? BEE Sigma is BEEBEE AI’s enterprise services brand: an AI consulting and engineering delivery team — and a manageable enterprise agent platform (the agents we deliver are often called AI digital employees). We plug a team of 12 specialized agents — three standard workflows plus custom Golden Flow builds — into the Teams, Slack, WeChat, Lark, email and ERP you already use, turning lead-to-cash into reusable, verifiable AI workflows and knowledge systems. Not another app to learn. ## Who is BEE Sigma for? Businesses whose follow-up, knowledge and delivery depend on human experience: we have delivered cases in education, construction, cross-border trade & logistics, and financial & business services. A simple test: if your team still copies from chats, pastes into spreadsheets and tracks clients from memory, it’s for you. Not sure whether to replace your business software first? Read our New Zealand business software comparison; most businesses don’t need to switch. ## How is this different from AI training? Training teaches tools; back at the desk, nothing changes. BEE Sigma is accountable for business results: starting from one concrete scenario, we work alongside you through process diagnosis, knowledge capture, agent workflow build-out and launch validation — with humans approving every high-risk action, and every step logged and reversible. ## What does a project typically deliver? Process diagnosis, knowledge asset organization, agent workflow design and configuration (starting from whichever is most valuable: GTM lead-gen, CRM follow-up or finance reconciliation), integration with your existing systems, team training, and a clear set of acceptance metrics. On-premise deployment supported — data stays with you. ## How soon do we see results? We recommend closing a minimal business loop in 2–4 weeks — most clients start with CRM follow-up or reconciliation, where results come fastest. Complex scenarios typically expand workflow by workflow over 3–6 months. Every step has acceptance criteria and a way back — you can stop any time. ## What do we need to contribute? Real business processes, historical cases, FAQs, delivery material, and interviews with key people — we turn that tacit experience into agent-usable knowledge. Expect 2–3 hours a week from the business owner in the early phase. ## What can it do for education businesses? Build an admissions funnel that never drops the thread, give every counsellor a closing co-pilot (pre-meeting match, live prompts, post-meeting profiles), turn senior counsellors’ experience into training for new hires, standardize parent communication, and make teaching and consulting SOPs reusable knowledge — with delivered education cases in both New Zealand and China. ## Where does BEE Sigma operate? We serve China, New Zealand, Australia, Thailand and Southeast Asia, in both English and Chinese. Our Auckland office is at B:Hive, Smales Farm (Takapuna), with team presence in Chengdu and Beijing. ## Which company is behind BEE Sigma, and is it legitimate? BEE Sigma is the enterprise services brand of BEEBEE AI, operated by BEEBEE AI Limited, registered with the New Zealand Companies Office under NZBN 9429051781781, with an office at B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland 0622 and a verifiable Google Business Profile. BEE Sigma is not a cryptocurrency, investment platform or trading tool and has no connection with any token of the same name; we only do business AI consulting and workflow implementation, and every published case is a dated, anonymised project record. ## What are real examples of AI technology applied in New Zealand? By BEE Sigma’s dated 2026 project records, AI applications in New Zealand businesses cluster in four workflows: supplier invoice routing (construction head contractor, 730 of 798 documents routed automatically in one accounting month, review of 1 Sep 2026), site timesheet entry (South Island builder, company-wide use after 10 weeks), cross-border orders to courier labels (health-products retailer, first production order after 90 days) and sales reconciliation (financial training firm, live in 11 days). No system was replaced; agents sit between Xero, ApprovalMax, Lark, WeChat and the ERP, and payment, approval and external commitments stay with people. Full records: AI implementation case studies and how businesses use AI to improve efficiency. ## BEE Sigma —how businesses enter the agent era Two free starting points, a few minutes each — see yourself clearly, then decide where to begin. AIV · AI VISIBILITY ## Free AIV Scan See how visible your site is in AI search — who gets recommended and why you don’t. The report also flags the process most worth transforming first. Start the scan AIM · ADOPTION ASSESSMENT ## Free AIM Assessment Answer a few questions about how you operate and see the best angle for adopting AI — which workflow to start with, and what to solve first. Start the assessment CONTACT · NEW ZEALAND & AUSTRALIAbrinny@beebee.ai CONTACT · MAINLAND CHINA, HK & TAIWANanna@2brain.ai ON-PREMISE · DATA STAYS WITH YOU · HUMAN-APPROVED ACTIONS · REGISTERED IN NEW ZEALAND BEESigma AI Native · The manageable enterprise agent platform.Agents plug into your existing systems; your team focuses on asking and deciding. Office: B:Hive, 74 Taharoto Road,Smales Farm, Takapuna, Auckland, New ZealandContact: Email us ## PRODUCT GTM Workflow ↗CRM Workflow ↗Finance Workflow ↗Golden Flow Custom ↗AIV Scan ↗AIM Assessment ↗ ## EXPLORE Blog ↗Case Studies ↗GTM Demo ↗CRM Demo ↗AI Consulting ↗AI Implementation ↗ ## COMPANY Founder ↗FAQ ↗Responsible AI ↗How We Work ↗NZ & AU · brinny@beebee.ai ↗Greater China · anna@2brain.ai ↗ © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · Google Business Profile 中文版 ===== # 新西兰 AI 咨询服务推荐 · 奥克兰落地陪跑团队 | BEE Sigma URL: https://beesigma.com/ai-consulting-auckland Summary: 总部位于奥克兰 B:Hive（Smales Farm）的 AI 咨询与落地陪跑团队：连接你已有的系统而不是替换，按工作流交付、按结果验收，中英双语，服务全新西兰。免费 AIV 诊断与 AIM 评估起步。 新西兰 AI 咨询服务推荐 · 奥克兰落地陪跑团队 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 AUCKLAND · AI CONSULTING & IMPLEMENTATION ## 新西兰 AI 咨询服务推荐：奥克兰的落地陪跑团队 我们不做 PPT 咨询：连接你已有的邮件、Teams、Slack、微信与 ERP，把获客、跟单、对账和你独有的业务流真正跑起来——按结果验收，上线后持续陪跑。 B:Hive, Smales Farm · Takapuna服务全新西兰 · 远程可达中英双语交付 免费 AIV 诊断AIM 落地评估 SERVICES · 我们做什么 ## 四条工作流，按结果验收 GTM ## 获客与培育 AI 可见度诊断与优化、市场雷达、内容生产、线索评分——让搜索引擎和 AI 推荐你，而不是竞品。 CRM ## 转化与签约 客户档案、提案辅助、跟进时间线、干净的合同交接——每一单推进得有记忆。 FIN ## 回款与合规 开票准备、银行流水自动核销、分级催款、税务申报准备——把现金流变成确定性。 GF ## Golden Flow 定制 你独有的产金流——工地工时、跨境订单、合规计算、行业专属流程，按你的业务模式定制。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） WHY · 为什么是我们 ## 连接、实干、陪跑——三句话说清差别 连接而不是替换： 一个系统都不用换，Agent 接入你已有的工具栈。 实干而不是论道： 写进合同的是验收口径，不是咨询报告；试点闭环通常 2–4 周。 陪跑而不是撤场： 上线只是开始——持续迭代、对业务结果负责。 本地在场也很重要：我们的办公室就在奥克兰 B:Hive（Smales Farm, Takapuna），熟悉 Privacy Act 2020 与 PIA 要求，新西兰本地已有教育、建筑、跨境零售等行业的交付案例；符合条件的企业还可申请 MBIE 最高 15,000 纽币的 AI 共同资助。 PATH · 怎么开始 ## 从免费诊断到长期陪跑 01免费 AIV 诊断看清官网在搜索与 AI 检索中的位置 02免费 AIM 评估找出最值得交给 Agent 的那条业务流 032–4 周试点固定范围报价，跑通最小业务闭环 04扩展与陪跑验证有效再逐条扩展，持续对结果负责 FAQ · 决定之前 ## 奥克兰企业常问的六个问题 ## 新西兰 AI 咨询服务推荐哪家？怎么判断靠不靠谱？ 先看三件事：是否要求你换系统（要求先换的要警惕）、交付物是工作流还是一份报告、上线后是否继续陪跑。新西兰的 AI 咨询服务商分三类：大型 SI 做政企大项目，单点工具工作室解决一个环节，落地陪跑型团队把 Agent 接进你已有的系统并按结果验收。BEE Sigma 属于第三类，奥克兰总部、服务全新西兰与中国，免费 AIV 诊断和 AIM 评估起步，报价在诊断之后。 ## 奥克兰有靠谱的 AI 咨询公司吗？怎么选？ 奥克兰聚集了新西兰大部分 AI 服务商，分三类：大型 SI（政企项目）、单点工具工作室、落地陪跑型团队。选择看三件事：是否要求换系统、是否敢写验收口径、交付后是否持续负责。BEE Sigma 属于落地陪跑型，总部就在奥克兰 B:Hive。 ## AI 咨询和落地要花多少钱？ 从免费开始：AIV 诊断与 AIM 评估都不收费，先看清值不值得做、从哪做。之后按工作流固定范围报价，试点闭环通常 2–4 周；符合条件的企业还可申请 MBIE 最高 15,000 纽币共同资助。 ## 多久能看到效果？ 跟单、对账、知识问答这类切入点通常 2–4 周跑通最小闭环；复杂场景 3–6 个月逐条工作流扩展，每一步有验收口径和回退机制。 ## 只服务奥克兰吗？ 不。总部在奥克兰，服务全新西兰（惠灵顿、基督城、汉密尔顿等均可远程交付），也服务中国与澳大利亚市场。 ## 团队讲中文可以吗？ 完全可以。我们中英双语交付：员工用中文提要求，客户看到英文的专业输出；工作流、培训与文档双语齐备。 ## 会不会做完就走？ 不会——这正是我们和外包开发的区别。上线后持续陪跑：监控运行、迭代规则、按业务结果负责，规则最终交给你的团队维护。 ## 把第一条业务流，交给 Agent 试试 先做一次免费 AIV 诊断，或花几分钟完成 AIM 评估——看清自己，再决定从哪里开始。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI 落地实践案例：新西兰 AI 技术落地实践案例 9 个（含中国项目，标注交付状态） | BEE Sigma URL: https://beesigma.com/cases Summary: 来自新西兰与中国两地的 9 个 AI 落地服务项目，每个标注交付状态（已上线、部分上线、试点 · 演示、开发中）：教育招生线索初筛、建筑应付发票、建筑公司应收账款、跨境零售微信群订单到顺丰面单、房贷顾问客户触达、园区招商对话式 CRM 等。企业名脱敏，数字取自标注日期的项目记录。 AI 落地实践案例：新西兰 AI 技术落地实践案例 9 个（含中国项目，标注交付状态） | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 CASEBOOK · 脱敏案例 ## 新西兰 AI 技术落地实践案例（含中国项目）：9 个 AI 落地实践案例，标注交付状态 来自新西兰与中国两地的服务项目。应客户要求隐去企业名称；每张卡片标注交付状态：已上线、部分上线、试点 · 演示或开发中，并写明原来的做法、Agent 现在负责的动作，以及人保留的决策。数字取自标注日期的项目复盘记录。 9 个服务项目 · 每个可点开看完整说明4 个行业 · NZ / CN / 跨境数字标注日期 · 企业名脱敏 CASEBOOK · 9 个案例 ## 九个服务项目，按交付状态如实标注 产品案例 · 老板分身已上线（产品） ## 老板分身：把老板的判断装进团队的 AI 智能体 产品案例 · BossClone · 新西兰与中国 6 家客户的使用记录 15–30 分钟六层访谈生成 SOUL.md，3 步装进飞书或企业微信里的智能体；含实操截图、流程图和客户群聊原话（已脱敏）。看完整案例：工作原理、实操与客户反馈 → 91.48%一个会计月发票自动分流（2026-09-01 复盘口径） 10 周工地工时从启动到全公司使用 11 天销售对账工作流立项到上线 90 天微信群订单到顺丰生产首单 9个服务项目 · 标注交付状态 2个市场（新西兰与中国） 教育与培训NZ试点 · 演示 ## 新线索进来，先有一封回信和预约链接 新西兰 · 国际教育集团 · 招生线索初筛（第一阶段试点） Lead 进入 CRM→自动首封邮件 + 预约链接→预约者进顾问队列 ✕ 原来两千多条新线索堆在 CRM 里（客户在 2026-07-31 会议上的估算），顾问只能逐个打电话，分不出优先级。 试点：新线索进入 Zoho 后自动发个性化邮件与预约链接，用“是否预约”筛出高意向交给顾问 Agent 已接入 Microsoft Teams；录取后未推进签证的学生由 Agent 标记并起草跟进邮件（测试环境验证）查看完整案例说明 → 教育与培训CN试点 · 演示 ## 一个学生的 18 份散落资料，变成一张可检索的成长档案 中国 · 留学咨询机构 · 成长教练服务档案化（方案共创阶段） 原始资料→Agent 逐份摘要→多维表格索引→教练按阶段调用 ✕ 原来测评报告、教练反馈、三年规划散在 PDF 和 Word 里，角色之间交接有断点。 首个学生样板：18 份原始资料加 10 份团队沉淀，建成 28 条结构化索引 服务方案与专业判断由机构提出并确认，AI 只负责结构化与表达查看完整案例说明 → 建筑与工程NZ已上线 ## 应付发票，不再靠人追着跑 新西兰 · 建筑总承包商 · 应付发票自动分流 发票识别→匹配 PO 与审批人→提交 ApprovalMax→状态回写仪表盘 ✕ 原来发票人工录入，审批卡在谁那里没人知道，逾期了才被发现。 Agent 识别字段、归档、匹配采购单并提交审批；审批与付款仍由人在 ApprovalMax 与 Xero 里完成 一个会计月 798 份文档中 730 份自动分流（91.48%，2026-09-01 项目复盘口径）；Admin 与审批人两版仪表盘完整实录：发票自动分流 91% → 建筑与工程NZ部分上线 ## 开票、催收、风控：Agent 起草，人最后一按 新西兰 · 建筑公司 · 应收账款自动化 报价单→Xero 开票并发送→每日拉取逾期→催款草稿三档→风险分级卡片 ✕ 原来开票靠人工录入，逾期账款靠财务翻 Xero 账龄表，催款邮件语气全凭当时心情。 Agent 读取 Google Drive 报价单，在 Xero 建发票并发送（2026-07 MVP 上线） 每天 09:00 拉取 Xero 逾期数据，生成三档语气的催款草稿，只生成不自动发送；四级风控卡片供管理层决策（2026-09 演示阶段）查看完整案例说明 → 建筑与工程NZ已上线 ## 工头一句话，工时进系统 新西兰 · 南岛建筑施工公司 · 工地工时与人工成本 飞书群一句话→Agent 解析写表→工头确认→按时薪自动计算→成本看板 ✕ 原来工时靠工头下班后回忆、再汇总进 Excel，老板看不到每个项目的人工成本。 工头在已有的飞书群里说一句话，Agent 写进日工时明细与工地日志；时薪计算走多维表格自动化，不经过大模型（2026-04-08 启动，06-15 全公司使用） 2026-06-27：20 个项目、334 条日常记录、累计 1 万多小时工时进入系统；工资与成本按角色分权限查看完整实录：工地工时从 Excel 到飞书群 → 跨境贸易与物流NZ ⇄ CN已上线 ## 微信群里的订单，第一次有了闭环 中新跨境 · 保健品零售 · 微信群订单到顺丰面单 群内订单→拆单与核验卡→人核对四项说“审核通过”→顺丰建单与证件归档→面单回调回写 ✕ 原来订单埋在群聊里，业务员要把同一笔订单抄进总表、顺丰模版和证件归档，再把面单一张张发回。 Agent 解析微信群订单、映射 SKU、拆单并生成核验卡；人核对收件、商品、付款、证件四项后下达“审核通过”，Agent 才调用顺丰接口建单（2026-09-02 生产首单） 顺丰出库后主动推送面单，Agent 归档并回写多维表格（2026-09-09 打通）；ERP 与微信均未更换完整实录：微信群下单到顺丰面单 → 跨境贸易与物流NZ ⇄ CN部分上线 ## 从建档到托书，一条货代售前流不再靠人盯 中新航线 · 国际货代 · 售前到目的港三条工作流（分阶段交付） CRM 建档→订单创建→报价 PDF 人工审批→确认邮件→托书生成 ✕ 原来订单创建、报价审批、托书交接没有统一规则，前期反复沟通产生大量无效订单。 售前流：建档、建单、报价审批、确认邮件、托书生成已串联，报价由 sales support 审批后才建单（2026-09 交付） 始发港 Pre-Alert 文件校验与目的港 ETA、放行提醒开发中，2026-09-15 阶段评审查看完整案例说明 → 金融与商业服务NZ部分上线 ## 从一封生日祝福邮件开始的老客户触达 新西兰 · 房贷顾问团队 · 内容运营与客户触达 企业邮箱授权→生日自动祝贺→小红书选题与出图→每日房贷市场简报 ✕ 原来老客户生日靠顾问自己想起来；小红书账号内容积累少，提炼不出风格。 客户生日祝贺邮件经 Outlook 授权后按日自动发出（2026-08 打通） 小红书选题池与出图规则建立，顾问与 Agent 单聊即可产出内容；每天 08:00 推送房贷、利率与 refix 政策简报查看完整案例说明 → 金融与商业服务CN试点 · 演示 ## 销售一句话，AI 填好 CRM 预览，人确认再落库 中国 · 园区与场馆招商 · 对话式 CRM 录入（MVP 演示阶段） 飞书群一句话→七项字段预览→回复“确认并写入”→客户 / 项目 / 跟进三表 ✕ 原来跟进记在聊天和脑子里，CRM 字段常年空着，历史数据散在各销售的 Excel。 销售在飞书群描述拜访情况，Agent 解析、查重并生成七项字段预览，精确回复“确认并写入”才落库（MVP v3，2026-09 演示） 联系人五类角色与关系分级，一键生成决策链图（2026-09-10 测试通过）查看完整案例说明 → 本页案例均来自 BEE Sigma 在新西兰与中国的真实服务项目，应客户要求隐去企业名称。为准确起见，每张卡片注明交付状态：已上线指工作流已在客户日常业务中运行并由客户团队使用；部分上线指部分环节已在用、其余在演示或开发中；试点 · 演示指已在测试环境或演示会上跑通、尚未全量投入日常使用；开发中指处于方案或开发阶段。所列数字均为记录中的项目复盘口径并标注日期，部分未经客户书面确认；项目效果因客户现有系统、数据与配合程度而异，不构成对其他企业的承诺。页面内容截至所标日期，之后系统由客户自行运营，我们不持有最新数据。 FAQ ## 看案例时，客户最常问的 ## 新西兰 AI 技术落地实践案例有哪些？效果怎么量化？ 这里是新西兰与中国两地 9 个服务项目，每个标注了交付状态：国际教育集团的招生线索初筛、建筑总承包商的应付发票分流、南岛建筑公司的工地工时、另一家建筑公司的应收账款、跨境零售的微信群订单到顺丰面单、房贷顾问的客户触达、园区招商的对话式 CRM 等。每个案例都写明原来的人工基线和现在的做法；验收口径在开工前写死，例如重复问题接住率、单据处理时长、人审通过率。 ## 案例为什么要脱敏？能看到完整版吗？ 应客户要求隐去企业名称，流程细节完整保留。预约免费诊断时说明所在行业，我们会提供对应行业的完整案例，包括具体指标。 ## 我们不在这些行业里，有参考价值吗？ 有。行业不同，但问题模式高度重复：数据散在群聊和邮箱、跟进靠个人记忆、对账靠人核。Agent 接入的是你已有的系统，工作流逻辑可以迁移，变的只是行业词汇。 ## 从这样的案例到我们自己上线，一般要多久？ 以已有记录为准：一条销售对账工作流 11 天上线，工地工时约 10 周全公司使用，发票分流约 3 个月出第一个完整会计月数字，跨境订单约 90 天到生产首单。周期取决于流程范围、系统接口与客户配合，每一步都有验收口径和回退机制，不满意可以停。 ## 把第一条业务流，交给 Agent 试试 想知道你的行业该从哪条流程切入？先做免费 AIM 落地评估，我们会带着对应行业的完整案例来谈。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI Implementation Case Studies in New Zealand and China: 9 Anonymised Projects | BEE Sigma URL: https://beesigma.com/cases-en Summary: Nine AI implementation engagements from New Zealand and China, each labelled with its delivery status (live, partly live, pilot/demo, in development):… AI Implementation Case Studies in New Zealand and China: 9 Anonymised Projects | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan CASEBOOK · ANONYMISED CASES ## AI Implementation Case Studies from New Zealand and China: Eight Anonymised Projects, Each with Its Delivery Status Nine engagements across 4 industries and 2 markets, New Zealand and China. Company names are withheld at clients’ request; each card states its delivery status (live, partly live, pilot/demo or in development), how the work used to be done, what the agent now handles and which decisions people keep. Figures come from dated project review records. 8 engagements · delivery status shown4 industries · NZ / CN / cross-borderDated figures · names withheld CASEBOOK · 9 CASES ## Nine engagements, each labelled with its delivery status PRODUCT CASE · BOSS CLONELive (product) ## Boss clone: the owner’s judgement inside the team’s AI agent Product case · BossClone · records from 6 clients in New Zealand and China A 15–30 minute six-layer interview generates SOUL.md, installed in a Lark or WeCom agent in 3 steps; with walkthrough screenshots, diagrams and anonymised client chats.Read the full case: how it works, walkthrough and client feedback → 91.48%invoices auto-routed in one accounting month (review of 1 Sep 2026) 10 wkssite timesheets from kickoff to company-wide use 11 dayssales reconciliation from kickoff to go-live 90 daysWeChat orders to first production courier order 9engagements · delivery status shown 2markets: NZ and China Education & TrainingNZPilot · Demo ## A first reply and a booking link within minutes of a new lead NZ · International education group · lead screening (phase-one pilot) Lead enters CRM→Auto first email + booking link→Booked leads to counsellor queue ✕ BeforeMore than two thousand new leads sat in the CRM (the client’s estimate at a 2026-07-31 meeting); counsellors could only call one by one with no way to prioritise. Pilot: new Zoho leads get a personalised email and booking link; those who book are routed to a counsellor Agent live in Microsoft Teams; admitted students with stalled visas are flagged with a draft follow-up (validated in a test environment)Read the full case → Education & TrainingCNPilot · Demo ## Eighteen scattered files, one searchable student record CN · Study-abroad consultancy · coaching-service documentation (co-design phase) Source files→Agent summaries→Lark Base index→Coach retrieval by stage ✕ BeforeAssessment reports, coach notes and three-year plans lived in PDFs and Word files with gaps at every hand-off. First pilot student: 18 source files plus 10 team notes indexed into 28 structured records Programme design and professional judgement stay with the firm; AI structures and draftsRead the full case → Construction & TradesNZLive ## Payables that no longer need chasing NZ · Construction head contractor · automatic invoice routing Invoice OCR→PO & approver match→Submit to ApprovalMax→Status write-back ✕ BeforeInvoices keyed by hand, approvals stuck with no visibility, overdue only discovered after the fact. Agents read, file, match purchase orders and submit approvals; people still approve in ApprovalMax and pay in Xero 730 of 798 documents routed automatically in one accounting month (91.48%, project review of 1 Sep 2026); admin and approver dashboardsFull case study: 91% invoice automation → Construction & TradesNZPartly live ## Invoicing, collections and risk: agents draft, people press send NZ · Construction company · receivables automation Quote→Xero invoice & send→Daily overdue pull→Reminder drafts, 3 tones→Risk tier cards ✕ BeforeInvoices were keyed by hand, overdue accounts meant reading the Xero ageing report, and reminder tone depended on the day. Agent reads quotes from Google Drive, creates and emails the Xero invoice (MVP live Jul 2026) Daily 09:00 pull of Xero overdue items into reminder drafts in three tones, drafts only, never auto-sent; four-tier risk cards for management (demo stage, Sep 2026)Read the full case → Construction & TradesNZLive ## One sentence from the foreman, hours in the system NZ · South Island builder · site hours and labour cost One sentence in Lark→Agent parses and writes→Foreman confirms→Pay rates applied automatically→Cost dashboard ✕ BeforeHours were recalled by foremen after work and totalled in Excel; the owner could not see labour cost per project. Foremen post one sentence in the Lark group they already use; the agent writes the daily hours detail and site diary, and pay rates are applied by a Base automation, not a language model (kickoff 2026-04-08, company-wide 06-15) 2026-06-27: 20 projects, 334 daily records and over 10,000 hours in the system; payroll and cost visible by roleFull case study: site timesheets from Excel to a chat group → Trade & LogisticsNZ ⇄ CNLive ## WeChat-group orders, closed-loop for the first time NZ ⇄ CN · Health-products retail · WeChat orders to SF Express labels Group order→Split & verification card→Human 4-point check→SF order & ID archive→Label callback ✕ BeforeOrders sat in group chats; one staff member copied each into a master sheet, the SF template and an ID archive, then sent labels back one by one. Agent parses WeChat orders, maps SKUs, splits and issues a verification card; a person checks recipient, product, payment and ID and says ‘review passed’ before the agent calls SF Express (first production order 2 Sep 2026) SF pushes the label after dispatch; the agent archives it and writes back to Lark Base (live 9 Sep 2026). ERP and WeChat unchangedFull case study: WeChat orders to SF Express labels → Trade & LogisticsNZ ⇄ CNPartly live ## From client file to shipping instruction, a pre-sales line nobody has to babysit NZ ⇄ CN lane · Freight forwarder · three workflows, pre-sales to destination port (phased delivery) CRM record→Order creation→Quote PDF approval→Confirmation email→Shipping instruction ✕ BeforeNo shared rules for order creation, quote approval or shipping-instruction hand-off; repeated back-and-forth produced many dead orders. Pre-sales: client record, order, quote approval, confirmation email and shipping instruction chained; quotes released only after sales-support approval (delivered Sep 2026) Origin-port pre-alert checks and destination-port ETA and release alerts in development; stage review 15 Sep 2026Read the full case → Finance & Business ServicesNZPartly live ## Client touchpoints that start with a birthday email NZ · Mortgage adviser team · content operations & client touch Mailbox authorised→Birthday emails→Xiaohongshu topics & images→Daily mortgage brief ✕ BeforeClient birthdays depended on an adviser remembering; the Xiaohongshu account had too little content to find a voice. Birthday emails go out automatically through the authorised Outlook mailbox (live Aug 2026) Topic pool and image rules set up; advisers produce posts by chatting with the agent; a daily 08:00 brief on mortgage rates, OCR and refix newsRead the full case → Finance & Business ServicesCNPilot · Demo ## One sentence from sales, a CRM preview from AI, saved only after confirmation CN · Venue & park leasing · conversational CRM capture (MVP demo stage) One sentence in Lark→Seven-field preview→Reply ‘confirm and write’→Client / project / follow-up tables ✕ BeforeFollow-ups lived in chats and heads; CRM fields sat empty; history was scattered across each salesperson’s Excel. Sales describe a visit in a Lark group; the agent parses, de-duplicates and previews seven fields; nothing is saved until they reply ‘confirm and write’ (MVP v3, demoed Sep 2026) Five contact roles and relationship tiers; one-click decision-chain diagram (tested 10 Sep 2026)Read the full case → The cases on this page come from BEE Sigma’s actual client engagements in New Zealand and China; company names are withheld at clients’ request. For accuracy, each card states its delivery status: Live means the workflow runs in the client’s day-to-day operations and is used by their team; Partly live means some steps are in use while others are in demo or development; Pilot · Demo means it has run in a test environment or demonstration but is not yet in full daily use; In development means it is at the design or build stage. Figures are taken from dated project review records and some have not been confirmed in writing by the client; results depend on each client’s existing systems, data and participation and are not a promise of outcomes for other businesses. Content is current as at the dates shown; systems are subsequently operated by clients and we do not hold later figures. FAQ ## What clients ask when reading the cases ## What AI implementation case studies do you have in New Zealand, and how are results measured? Nine engagements across New Zealand and China, each labelled with its delivery status: lead screening for an international education group, payables routing for a construction head contractor, site timesheets for a South Island builder, receivables automation for another construction company, WeChat orders to SF Express labels for a cross-border retailer, client touchpoints for mortgage advisers, and conversational CRM capture for a venue-leasing operator. Each case states the manual baseline and what changed; acceptance metrics such as repeat-question absorption, document turnaround and human-review pass rate are written before work starts. ## Why are the cases anonymised, and can we see full versions? Clients asked us to withhold company names while keeping the full process detail. Tell us your industry when booking a diagnosis and we’ll walk you through the complete case for that sector, including the metrics. ## We’re not in one of these industries. Is this still relevant? Yes. The industries differ, but the patterns repeat: data scattered across chats and inboxes, follow-up that depends on memory, reconciliation done by hand. The agents connect to the systems you already use, so the workflow logic transfers; only the vocabulary changes. ## How long from a case like this to our own go-live? Going by our records: a sales reconciliation workflow went live in 11 days, site timesheets reached company-wide use in about 10 weeks, invoice routing produced its first full accounting-month figure after about 3 months, and cross-border ordering took about 90 days to the first production order. Timing depends on scope, interfaces and client participation; every step has acceptance criteria and a rollback, and you can stop if unsatisfied. ## Hand your first workflow to the agents Want to know which workflow your industry should start with? Take the free AIM assessment and we’ll bring the full case for your sector. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰企业智能化转型服务：从诊断到 Agent 工作流上线 | BEE Sigma URL: https://beesigma.com/enterprise-ai-transformation-nz Summary: BEE Sigma 面向新西兰与中国企业的智能化转型服务：不换系统，先做免费诊断与落地评估，再选一条最值得的业务流程做 2–4 周试点，把 12 个职能 Agent 接入你已有的邮件、Teams、Slack、微信、飞书与 ERP，按结果验收、上线后持续陪跑。奥克兰总部，中英双语交付。 新西兰企业智能化转型服务：从诊断到 Agent 工作流上线 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 NEW ZEALAND · ENTERPRISE AI TRANSFORMATION ## 新西兰企业智能化转型服务，从一条真正产金的流程开始 智能化转型不是换一套系统，也不是全员上一门 AI 课。我们先诊断你的流程和数据现状，再挑一条最值得的业务流做试点：机器跑动作，人做决策，四周内用可核对的数字说话。 免费诊断 + 落地评估起步不换系统 · 接入现有工具栈奥克兰总部 · 服务全新西兰与中国 免费 AIV 诊断AIM 落地评估 ROADMAP · 转型四步 ## 四步走完，转型才算落地 01 ## 诊断：看外也看内 AIV 可见度诊断看客户和大模型怎么找到你；AIM 落地评估盘清内部哪条流程重复度高、规则清晰、数据现成、错了能兜住。两项免费，做完拿到明确切入点。 02 ## 试点：一条流程，2–4 周 从 GTM 获客、CRM 跟单或财务对账里挑价值最高的一条，接入你现有的邮件、Teams、Slack、微信、飞书与 ERP。开工前写死验收口径：基线、一到两个硬指标、人审通过率。 03 ## 扩展：三条标准流逐条铺开 试点达标后，按获客与培育、转化与签约、回款与合规三条标准流扩展。12 个职能 Agent 分工明确，涉钱、涉合同、对外承诺的动作一律留人审确认点。 04 ## Golden Flow：长出你独有的产金流 真正含金量最高的流程在你的业务模式里：工地工时记录（10 周全公司使用）、微信群订单到顺丰面单（90 天生产首单）、销售对账（11 天上线）都是有记录的 Golden Flow 实例。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） 我们在新西兰企业 AI 落地指南里把三类供应商拆开讲过：大型 SI 适合政企大项目，单点工具解决一个环节，落地陪跑型团队把 AI 接进你已有的系统、陪到流程真正跑起来。BEE Sigma 是第三类：连接而不是替换，实干而不是论道，陪跑而不是撤场。符合条件的新西兰企业还可以申请 MBIE AI Advisory Pilot 补贴，最高 15,000 纽币。 FAQ ## 关于企业智能化转型，新西兰老板最常问的 ## 新西兰企业智能化转型服务包括什么？和买一套 AI 软件有什么区别？ 包括四步：免费诊断（AI 可见度 + 落地评估）、2–4 周单流程试点、三条标准流扩展、Golden Flow 定制。区别在于我们不卖软件：12 个职能 Agent 接进你已有的邮件、Teams、Slack、微信、飞书与 ERP，按工作流交付、按结果验收，上线后继续陪跑。买软件买到的是工具，转型服务买到的是跑起来的流程。 ## 中小企业做智能化转型要多少钱、多久见效？ 诊断和评估免费。之后按工作流固定范围报价，我们不在页面标价，正式报价在免费诊断之后给出。周期上，选对切入环节的试点 2–4 周应看到可核对的数字（重复问题接住率、单据处理时长等）；复杂场景 3–6 个月逐条工作流扩展。符合条件的新西兰企业可申请 MBIE AI Advisory Pilot，最高 15,000 纽币的 50% 共同资助。 ## 智能化转型需要换掉现有的 ERP、CRM 或 Xero 吗？ 不需要，多数情况下也不应该。换系统意味着停摆、迁移和全员重新培训。我们的口径是连接而不是替换：Agent 接进你已经在用的系统，接管抄录、催办、核对这类重复动作，报价、成交、付款等决策留在人手里。 ## 先从哪条流程开始转型最稳？ 符合“高频、规则清晰、数据现成、错了能兜住”四个条件的流程：线索没人跟先做 GTM，单子丢在群聊里先做 CRM 跟单，账收不回来先做财务对账与回款。重大谈判、高判断含量的环节不要当第一个试点。 ## 数据安全和新西兰 Privacy Act 合规怎么保证？ 开工前先做目的限定和数据最小化：哪些数据可以不收、去标识或留在你的边界内，白纸黑字写清；Agent 只在授权范围内读写；涉及个人信息的高影响决定设人审确认点，并保留日志与回退路径。具体控制取决于项目风险与合同范围，详见我们的负责任 AI 页面。 ## BEE Sigma 在哪里？只服务奥克兰吗？ 总部在奥克兰 B:Hive（Smales Farm, Takapuna），服务全新西兰，惠灵顿、基督城、汉密尔顿均可远程交付；同时服务中国与澳大利亚市场，中英双语团队。3EYES、华樱教育、财迷游言等都是服务中的客户。 ## 把第一条业务流，交给 Agent 试试 从两个免费入口开始：AIV 诊断看你在 AI 搜索里的位置，AIM 评估找出最值得先转的那条流程。正式报价在免费诊断之后，不在页面标价。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Enterprise AI Transformation Services in New Zealand: From Diagnosis to Live Agent Workflows | BEE Sigma URL: https://beesigma.com/enterprise-ai-transformation-nz-en Summary: Enterprise AI transformation services for New Zealand and China businesses: free diagnosis and assessment first, then a 2–4 week pilot on one workflow,… Enterprise AI Transformation Services in New Zealand: From Diagnosis to Live Agent Workflows | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan NEW ZEALAND · ENTERPRISE AI TRANSFORMATION ## Enterprise AI Transformation Services in New Zealand, Starting with One Workflow That Pays AI transformation is not a system swap, and it is not an AI course for the whole staff. We diagnose your processes and data first, pick the one workflow most worth automating, and run a pilot where machines do the motions and people keep the decisions — with checkable numbers inside four weeks. Free diagnosis + assessment to startNo system swap · plugs into your stackAuckland HQ · all of NZ and China Free AIV ScanAIM Assessment ROADMAP · FOUR STEPS ## Four steps, and transformation actually lands 01 ## Diagnose: outward and inward The AIV visibility scan shows how customers and large language models find you; the AIM assessment maps which internal workflow is high-frequency, rule-clear, data-ready and mistake-recoverable. Both are free, and you leave with a concrete entry point. 02 ## Pilot: one workflow, 2–4 weeks Pick the highest-value flow among GTM lead-gen, CRM follow-up and finance reconciliation, and plug agents into the email, Teams, Slack, WeChat, Lark and ERP you already use. Acceptance criteria are written before work starts: a baseline, one or two hard metrics, and the human-review pass rate. 03 ## Scale: three standard workflows Once the pilot clears its numbers, expand across acquisition and nurturing, conversion and signing, collections and compliance. Twelve role agents with clear jobs, and a human checkpoint on every payment, contract and external commitment. 04 ## Golden Flow: grow the flow only you have The highest-value workflow lives inside your business model. Site timesheets (company-wide in 10 weeks), WeChat orders to SF Express labels (first production order in 90 days) and sales reconciliation (live in 11 days) are all recorded Golden Flows. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) In our guide to AI implementation in New Zealand we split providers into three kinds: large SIs for government and corporate programmes, single-point tools for one step, and hands-on implementation partners who connect AI into the systems you already run and stay until the workflow is live. BEE Sigma is the third kind: connect don't replace, build don't just advise, stay don't ship-and-leave. Eligible NZ businesses can also apply for the MBIE AI Advisory Pilot, up to NZD 15,000. FAQ ## What New Zealand business owners ask about AI transformation ## What do enterprise AI transformation services in New Zealand include, and how is that different from buying AI software? Four steps: a free diagnosis (AI visibility plus implementation assessment), a 2–4 week single-workflow pilot, expansion across three standard workflows, and Golden Flow custom builds. The difference is that we don't sell software: 12 role agents plug into the email, Teams, Slack, WeChat, Lark and ERP you already run, delivery is by workflow and accepted by results, and we stay after launch. Buying software gets you a tool; a transformation service gets you a running process. ## How much does AI transformation cost for an SME, and how soon do we see results? Diagnosis and assessment are free. Delivery is then priced per workflow with a fixed scope; we publish no prices, and the formal quote follows the free diagnosis. On timing, a well-chosen pilot shows checkable numbers within 2–4 weeks (repeat-question absorption, document turnaround); complex scenarios expand workflow by workflow over 3–6 months. Eligible NZ businesses can apply for the MBIE AI Advisory Pilot, which co-funds 50% up to NZD 15,000. ## Do we have to replace our ERP, CRM or Xero to transform with AI? No, and in most cases you shouldn't. Replacement means downtime, migration and retraining everyone. Our rule is connect, don't replace: agents plug into the systems you already use and take over the repetition (transcribing, chasing, matching) while pricing, deals and payments stay human decisions. ## Which workflow is the safest place to start? One that is high-frequency, rule-clear, data-ready and mistake-recoverable. Leads going cold: start with GTM. Deals buried in group chats: start with CRM follow-up. Cash stuck in receivables: start with reconciliation and collections. Major negotiations and judgment-heavy work should never be the first pilot. ## How do you handle data security and New Zealand Privacy Act compliance? Purpose limitation and data minimisation come first: what can be left uncollected, de-identified or kept inside your boundary is agreed in writing. Agents read and write only within granted scope; high-impact decisions involving personal information get a human checkpoint, with logs and rollback paths. Controls scale with project risk and contract scope; see our Responsible AI page. ## Where is BEE Sigma based, and do you only serve Auckland? Headquartered at B:Hive, Smales Farm, Takapuna in Auckland, we serve all of New Zealand (Wellington, Christchurch and Hamilton remotely) plus China and Australia, with a bilingual English–Chinese team. 3EYES, 华樱教育 and 财迷游言 are among current clients. ## Hand your first workflow to the agents Start with two free entry points: the AIV scan shows where you stand in AI search, and the AIM assessment finds the one workflow worth transforming first. The formal quote comes after the free diagnosis, never from a price list. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 技术落地实践案例：一家跨境保健品零售企业，从微信群下单到顺丰面单的一百天 | BEE Sigma URL: https://beesigma.com/insight-case-nz-cross-border-wechat-orders-sf-express Summary: 新西兰 AI 技术落地实践案例（跨境电商）：一家在新西兰开着 5 家门店、在中国走十几个电商渠道的保健品零售企业。合作前，国内日销靠截图和 Excel 逐条录入，顺丰下单靠业务员把微信订单抄进总表再复制进模版。一百天后，客户在微信群里发一段话，智能体拆单、核验、建单、归档证件，顺丰把面单主动推回来。 新西兰 AI 技术落地实践案例：一家跨境保健品零售企业，从微信群下单到顺丰面单的一百天 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · AI 落地案例 · 新西兰跨境零售 ## 新西兰 AI 技术落地实践案例：一家跨境保健品零售企业，从微信群下单到顺丰面单的一百天 内容复核 2026.09.148 分钟BEE Sigma 落地团队 这篇新西兰 AI 落地案例的主角是一家跨境保健品零售企业：新西兰 5 家门店和一个本地网店，中国侧走天猫、京东、拼多多、抖音等十余个渠道，卖奶粉、保健品和蜂蜜，国际物流以顺丰为主。合作前，国内日销靠人看截图、逐条录进 Excel；顺丰下单靠一位业务员把微信群里的订单先抄进总表，再复制到下单模版导入系统。2026 年 6 月 4 日启动，9 月 9 日顺丰开始主动推送面单。下面用图讲这一百天。 01 ## 原来一笔订单要抄几遍？ 7 月 24 日，客户方负责顺丰下单的业务员这样描述自己的工作：“微信订单下做到总表里记录各种信息用的，然后复制内容到下单模版，导入顺丰系统。”一笔奶粉订单从微信群到顺丰面单，要经过人手五次。 合作前后对照。右侧唯一需要人做的，是核对四项后说一句“审核通过”。 02 ## 一百天发生了什么 6 月 4 日启动，先做三个数据场景；7 月底因为电商平台 API 的合规规则换路线，转做顺丰下单；8 月 26 日顺丰接口接通；9 月 9 日面单回调打通；9 月 10 日转入陪跑。 八个节点，每个都有群聊记录可查。 03 ## 第一个月：先做最小的事 第一阶段只做三件事：群消息检索、国内日销报表、门店日报。四周一个闭环，第二周就拉客户进测试群。7 月 9 日首月复盘时定下一条后来反复被验证的共识：第三方接口只是原子能力，“不存在有接口即可直接使用的情况”。 6 月 9 日的首月计划。目标写得很朴素：快速让客户感受机器人作用，解决眼前数据问题。7 月 9 日首月复盘：四个场景的状态，以及每个阶段“计划、实际、差距”的对照。 04 ## 两道坎：微信号与平台 API 让智能体读微信群，微信号得长期登录在一台在线的机器上。微信对异地和多端登录风控很严，双方配合了近一个月，8 月 5 日专用号才稳定接入。这段经历变成了我们所有微信项目启动清单的第一条。 客户希望直接接淘宝、视频号的 API。研究一周后发现平台要求申请方持有与店铺主体一致的软件著作权，本地部署的智能体拿不到。7 月 28 日换路线：国内日销保持“人报数、机器人记”，开发资源转向顺丰下单。 一个月的拉锯换来的四条前置条件。7 月 21 日到 28 日的判断：一周研究清楚，然后换一条路。 05 ## 一笔订单的旅程 8 月 28 日测试环境首单，9 月 2 日生产环境首单。整条链路里只有一个节点必须由人出手：核对收件信息、商品、付款、证件四项后，对智能体说“审核通过”。没有这句话，智能体不会调用顺丰的生产接口。 八个节点，金色的那个是人。8 月 31 日的一次真实往返：人说“核验通过”，智能体只剩一个问题“6 袋是不是 1 个履约单元”，人答“一单 6 袋”，它才出单。人名与单号已打码。8 月 26 日顺丰接口接通当天：运单号拿到了，面单没有。面单这件事接下来卡了两周。 06 ## 面单为什么拿不到 对跨境电商，面单是给客户的收据，拿不到等于没做完。顺丰的机制是仓库拣货装箱、出库审核通过后才生成面单，出库前接口查不到，网页能看到也查不到。最终方案是顺丰主动回调：9 月 9 日晚生产环境打通，9 月 9 日面单回调打通后，面单路径回写到工单。 建单和拿面单是两个独立动作。8 月 31 日顺丰对接人的解释：面单以实际打包出库为准。群名、人名与头像已打码。 07 ## 一次“假失败” 9 月 2 日晚，一单数量从 1 改成 6，取消重建后接口报“无库存”，工单被标记失败。智能体自查发现重建时误换了 SKU，回查顺丰：单其实已建好、运单已生成。它的结论写进了工作流规则：“请勿再次重试或再建单，避免重复发货。” 系统报错不等于顺丰没建单。任何报错先用交易号回查对方系统，再决定是否重试。 08 ## 一个改错了一个月的公式 7 月 30 日，客户希望利润计算的成本口径从“平均成本”改成“最新成本”，我们改了公式。日报每天 21 点准时发出、格式完整，没有一个环节报错，利润率却在 74% 到 96% 之间停了一个月。8 月 29 日客户看数据时提出偏高，9 月 1 日修复。教训在我们：改核心公式时没有同时设“超出正常区间就报警”。 日报显示的利润率，数值取自当日日报卡片。8 月 29 日当天的日报。表格完整、准时，正是它看起来正常，才一个月没人看第二眼。门店名与金额已打码。 09 ## 机器人为什么会“忘事” 对话越长，早期约定的规则越可能在自动压缩时被丢掉。“不是忘了，而是已经看不到了。”所以规则要写进技能文件和多维表格，不能只留在聊天里。这张 6 月 12 日发给客户的图解，后来成了我们所有项目的默认说明。 上下文压缩机制图解。解法是把规则写进文件。 10 ## 合作方式：第一季度做一条，之后教着做 第一季度全托管，搭建、调试、对接由我们完成；第二到第四季度陪跑，我们出方案、答疑、复盘，落地由客户团队自己完成。目的是让客户长出自己用 AI 的能力。顺丰下单是客户说清“真正占人力的环节还没释放”后加进第一季度做完的，这就是“先做一条、做透一条”。9 月 10 日起，我们做教练。 7 月 9 日首月复盘纪要里的服务边界表。 11 ## 五条教训 回头看，最难的部分没有一件是模型能力。 前置条件先于开发：合规的专用微信号要提前一个月准备，写进启动清单第一条。 合规卡点要早判断：一周研究清楚然后换路，比拖三周好。 建单和拿单据是两件事：面单在出库后才生成，靠对方回调，不靠我们轮询。 系统报错不等于事情没发生：任何报错先用交易号回查，再决定重试。 核心公式改动要有阈值校验：改公式的同时设报警，这是我们自己该做到的。 12 ## 现在怎样了？ 9 月 11 日，微信下单、核验、出单、证件归档、面单回传的日常工单在正常跑。客户方运营人员开始提下一步：让机器人每一两小时出一份下单总表，面单批量下载；十几个客户每天要更新的账单表也想交给机器人。国内日销和门店日报每天 21 点准时发出，利润率回到正常区间。 证据边界：本文数字与引文来自项目群聊记录和 7 月 9 日、9 月 1 日两份会议纪要，日期均已标注。门店与渠道的日销售额不写具体数值；顺丰交易号、订单号、运单号、证件与收件人信息一律未采用。客户为新西兰跨境保健品零售企业，已匿名；截图中的公司名、人名、头像与单号已裁除或打码；示意图为本文绘制。 ## 一手参考 顺丰速运（官网） ↗飞书多维表格（官方产品页） ↗ FAQ · 快速回答 ## 新西兰跨境电商的 AI 落地案例里，微信群下单能自动生成顺丰面单吗？ 能，分两步。智能体把微信群里的自然语言订单拆单、核验、调用顺丰 ISCO 接口创建出库单，几分钟完成；面单要等顺丰仓库拣货装箱、出库审核通过后才生成，由顺丰通过回调主动推送，智能体再归档并回写表格。“下单即出面单”在顺丰的机制下做不到。 ## 这个案例里 AI 做哪一步，人做哪一步？ AI 做：解析订单、映射 SKU、拆单、生成核验卡片、归档证件、调用顺丰接口建单、回写单号和面单、汇总日销与门店日报。人做：核对收件信息、商品、付款、证件四项后说“审核通过”；判断是否需要清关证件；报错时决定是否重试。没有人工指令，智能体不调用顺丰生产接口。 ## 需要换 ERP 或换微信、飞书吗？ 不需要。客户的 ERP、顺丰 ISCO、微信、群晖 NAS 一个都没换，智能体跑在客户自己的 Mac mini 上，用 OpenClaw 做调度，数据落在客户自己的飞书多维表格里。唯一新增的是一个长期在国内登录的专用微信号。 ## 微信下单到顺丰出单多久能跑通？ 这个项目 7 月 31 日确定顺丰范围，8 月 26 日接口接通，8 月 28 日测试环境跑通，9 月 2 日生产首单，9 月 9 日面单回调打通，约 6 周。开发本身不到一半，其余时间花在顺丰正式环境参数、面单机制确认和微信号接入上。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # New Zealand AI Implementation Case Study: From WeChat Orders to SF Express Labels at a Cross-Border Health Retailer | BEE Sigma URL: https://beesigma.com/insight-case-nz-cross-border-wechat-orders-sf-express-en Summary: AI implementation case study, New Zealand (WeChat, Lark Base, OpenClaw, SF Express ISCO): a health-products retailer with 5 stores in New Zealand and a… New Zealand AI Implementation Case Study: From WeChat Orders to SF Express Labels at a Cross-Border Health Retailer | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · AI Implementation Case Study · NZ Cross-Border Retail ## New Zealand AI Implementation Case Study: From WeChat Orders to SF Express Labels at a Cross-Border Health Retailer Reviewed 14 Sep 20268 minBEE Sigma Delivery This New Zealand AI implementation case study is about a cross-border health-products retailer: 5 stores and an online shop in New Zealand, and more than ten channels in China including Tmall, JD, Pinduoduo and Douyin, selling infant formula, supplements and honey, with SF Express as the main carrier. Before we started, daily China sales were read off screenshots and typed into Excel; SF orders were copied by one staff member from WeChat into a master sheet, then into an import template. The project started on 4 June 2026; on 9 September SF Express began pushing labels back automatically. Here are the hundred days, mostly in pictures. 01 ## How many times was one order copied? On 24 July the client’s SF operator described her routine: ‘WeChat orders go into my master sheet, then I copy the content into the order template and import it into the SF system.’ One formula order passed through human hands five times between the WeChat group and the label. Before and after. On the right, the only human step is checking four items and saying ‘review passed’. 02 ## What happened in a hundred days Kickoff on 4 June with three data scenarios; a change of route at the end of July when marketplace APIs ran into copyright rules, toward SF Express ordering; SF API connected on 26 August; label callback live on 9 September; coaching mode from 10 September. Eight milestones, each traceable in the project chat. 03 ## Month one: the smallest thing first Phase one was three things: group message search, a China daily sales report, a store daily report. One loop in four weeks, with the client in a test group from week two. The first-month review on 9 July recorded an agreement that kept proving true: third-party interfaces are atomic capabilities, and ‘there is no such thing as “the interface exists, so it can be used directly”’. The first-month plan of 9 June. The stated goal: let the client feel what the bots do quickly and fix the data problems in front of them. (Chinese-language original.)First-month review, 9 July: the state of the four scenarios and a plan-versus-actual table. (Chinese-language original.) 04 ## Two hurdles: the WeChat account and marketplace APIs For an agent to read WeChat groups, an account must stay logged in on an always-on machine. WeChat’s risk controls on remote and multi-device logins are strict; it took nearly a month of joint effort before the dedicated account was stable on 5 August. That experience is now item one on our kickoff checklist for every WeChat project. The client wanted the agent on Taobao and WeChat Channels APIs directly. A week of research showed the platforms require a software copyright held by the store entity, which a locally deployed agent cannot obtain. On 28 July the route changed: daily sales stayed in ‘people report, the bot records’ mode, and development moved to SF Express ordering. Four preconditions, earned over a month.The decision of 21 to 28 July: a week of research, then a different road. 05 ## The journey of one order First test-environment order on 28 August, first production order on 2 September. Only one node in the chain requires a person: check recipient, product, payment and ID, then tell the agent ‘review passed’. Without that sentence the agent does not call SF’s production API. Eight nodes. The gold one is a person.A real round trip on 31 August: a person says ‘verification passed’, the agent has one question left, the person answers, and only then does it create the order. Names and numbers pixelated. (Chinese-language original.)26 August, the day the SF API connected: waybill number yes, label no. The label took another two weeks. (Chinese-language original.) 06 ## Why the label would not come For a cross-border seller the label is the customer’s receipt; without it the order is not finished. SF Express generates the label only after the warehouse picks and packs and the outbound order is approved; before that the API returns nothing, even when the web page shows it. The final answer was a callback from SF: live in production on the evening of 9 September, with label paths written back to work orders once the 9 September callback was live. Creating an order and getting the label are two separate actions.SF’s integration contact on 31 August: the label follows the actual pack-out. Group name, names and avatars pixelated. (Chinese-language original.) 07 ## A ‘false failure’ On the evening of 2 September an order’s quantity changed from 1 to 6; after cancel-and-rebuild the API reported ‘no stock’ and the work order was marked failed. The agent checked itself, found the rebuild had switched SKU, and queried SF: the order in fact existed with a waybill. Its conclusion went into the workflow rules: ‘Do not retry or create another order, to avoid shipping twice.’ A system error does not mean SF did not create the order. On any error, query the other system by transaction number first. 08 ## A formula that was wrong for a month On 30 July the client wanted the cost basis in the profit calculation changed from average to latest cost, and we changed the formula. The report went out at 21:00 every day, complete, with no step reporting an error, while the margin sat between 74% and 96% for a month. The client flagged it on 29 August; we fixed it on 1 September. The lesson is ours: a change to a core formula needs an ‘outside the normal range’ alert in the same commit. The margin shown on the daily report; values from the report cards.The report of 29 August. Complete and on time, which is exactly why nobody looked twice for a month. Store names and amounts pixelated. 09 ## Why does the bot ‘forget’? The longer a conversation, the more likely early rules are dropped in automatic compression. ‘It did not forget. It can no longer see it.’ So rules go into skill files and Base tables, never only into chat. This diagram, sent to the client on 12 June, became our default explanation on every project. How context compression works. The fix is to write rules into files. (Chinese-language original.) 10 ## The engagement: build one line in Q1, then coach The first quarter is fully managed: we build, tune and integrate. Quarters two to four are coaching: we provide plans, answers and reviews, and the client’s team implements. The aim is for the client to grow its own AI capability. SF ordering was added to the first quarter after the client made clear that the step consuming staff hours had not yet been freed; that is what ‘build one line and build it through’ means. From 10 September, we coach. The service-boundary table from the 9 July review notes. (Chinese-language original.) 11 ## Five lessons Looking back, not one of the hardest parts was model capability. Preconditions before development: A compliant dedicated WeChat account, prepared a month ahead, at the top of the kickoff checklist. Judge compliance blockers early: A week of research and a change of road beats three weeks of drift. Creating an order and getting the document are two things: The label exists only after outbound; rely on the carrier’s callback, not our polling. A system error does not mean nothing happened: On any error, query by transaction number first, then decide whether to retry. Core formula changes need threshold alerts: Set the alert in the same change. That part is on us. 12 ## Where is it now? On 11 September the daily work orders for WeChat ordering, verification, SF order creation, ID archiving and label return were running normally, and the client’s operator was asking for the next step: an order summary every hour or two with batch label download, and the dozen customer statements she updates daily. China daily sales and the store report go out at 21:00 every day, with margins back in the normal range. Evidence boundary: numbers and quotations come from the project group chat and the meeting notes of 9 July and 1 September, each dated in the text. Daily sales figures for stores and channels are not given; SF transaction, order and waybill numbers, ID documents and recipient details were not used. The client is a New Zealand cross-border health-products retailer and is anonymised; company names, personal names, avatars and numbers in screenshots are cropped or pixelated; diagrams were drawn for this article. ## Primary sources SF Express (official site) ↗Lark Base (official product page) ↗ FAQ · Quick answers ## In this New Zealand cross-border e-commerce AI case study, can a WeChat order automatically produce an SF Express label? Yes, in two steps. The agent splits and verifies the plain-language order and calls the SF Express ISCO API to create the outbound order within minutes. The label is generated only after the SF warehouse picks and packs and the outbound order is approved; SF then pushes it through a callback, and the agent archives it and writes it back. ‘Label at the moment of ordering’ is not possible under SF’s mechanism. ## What does the AI do in this case, and what do people do? The AI parses orders, maps SKUs, splits by carton rule, issues verification cards, archives IDs, calls the SF API, writes back numbers and labels, and compiles daily sales and store reports. People check recipient, product, payment and ID and say ‘review passed’; decide whether customs documents are needed; and decide whether to retry on an error. Without a human instruction the agent does not call the production API. ## Do we need to replace the ERP, or move off WeChat and Lark? No. The client’s ERP, SF Express ISCO, WeChat and Synology NAS were all kept. Agents run on the client’s own Mac mini, orchestrated by OpenClaw, and data lands in the client’s own Lark Base. The only addition is a dedicated WeChat account kept logged in inside China. ## How long does WeChat-to-SF-Express ordering take to get running? SF scope fixed 31 July, API connected 26 August, full test flow 28 August, first production order 2 September, label callback 9 September: about six weeks. Development was less than half; the rest went on SF production parameters, confirming the label mechanism and WeChat account access. DT BEE Sigma Delivery The front-line team that plugs workflows into the systems businesses already run — methods drawn from delivered projects. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 技术落地实践案例：建筑公司工地工时管理三个月 | BEE Sigma URL: https://beesigma.com/insight-case-nz-construction-site-timesheet-ai Summary: 新西兰 AI 技术落地实践案例（建筑业）：一家南岛建筑施工公司，工时靠工头下班后回忆、Excel 汇总，项目信息散在微信群和邮件里。三个月后，工头在飞书群里说一句话，智能体写进日工时明细，老板在成本看板上看到 20 个项目、1 万多小时工时、334 条日常记录。 新西兰 AI 技术落地实践案例：建筑公司工地工时管理三个月 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · AI 落地案例 · 新西兰建筑 ## 新西兰 AI 技术落地实践案例：一家建筑公司的工地工时管理，从 Excel 到飞书群与成本看板的三个月 内容复核 2026.09.148 分钟BEE Sigma 落地团队 这篇新西兰 AI 落地案例的主角是南岛一家建筑施工公司，既做总包也做木工分包，几十名工人分布在近二十个项目上。合作前，工时靠工头下班后回忆和 Excel 汇总，老板看不到各项目的人工成本。2026 年 4 月 8 日启动，5 月 7 日第一版跑通，6 月 15 日全公司使用。6 月 27 日一季度交付汇报：12 个核心表格、20 个项目、334 条日常记录、累计 1 万多小时工时进了飞书多维表格。下面用图讲中间三个月发生了什么。 01 ## 原来，工时是怎么记的？ 公司内部沟通在微信群，对外在邮件，工时在 Excel：工头下班后回忆当天谁干了几个小时，月底会计汇总算工资。4 月 9 日老板提出需求：一套工地工人的工时、项目、每日填写与追踪系统，工地管理、工人、会计和行政都能快速看到自己需要的信息；4 月 28 日又补了 VO（变更单）和工地日志。拆开是三件事：谁去哪个工地干了多久、每个工地每天发生了什么、每个项目到今天花了多少人工，中间没有一张连着的表。 合作前后工时的记法对照。左边靠回忆和 Excel，右边靠群里一句话。 02 ## 第一刀切在哪里：先做工时，不做整个系统 4 月 21 日的拆解会上，我们把第一阶段定为工时统计：数据量最大、每天都发生、最容易验证对错；项目进度管理放第二阶段。两条原则当天定下：工头填的东西尽量下拉选择和智能判断，不在手机上自由打字；语音只用来调出表格，不直接下命令。4 月 22 日方案画成五步流程和四个角色，工人不需要公司账号，只在一个 Lark 外部群里收通知。 4 月 22 日第一阶段方案里的五步流程：早上发任务、工头确认修改、VO 按需触发、下班前汇报实际工时、系统汇总生成工地日志。同一份方案里的四个角色。工头看不到财务成本，财务通过独立私密表看工时与成本，时薪表和工时表分离。 03 ## 我们接进了什么？ 原则仍然是连接而不是替换：工人还是在群里说话，老板还是看表，中间多了一层会写表的智能体。每个工地一个飞书群，工头把完成情况、突发事件、照片发进群并 @ 智能体；智能体跑在客户办公室的 Mac mini 上，用 OpenClaw 调度，工时、工地日志、VO 各一个智能体。底座是一套多维表格：日工时明细表是工时和人工成本的唯一来源，工时到工资由多维表格自动化按时薪计算，不走大模型。智能体不改时薪、不算工资单、不批任何东西。 一句话到成本看板的数据流。金色节点是人；工资计算不走大模型。6 月 1 日部署前的数据表清单，以及“施工项目总览｜成本驾驭舱”仪表盘。4 月 23 日搭的“工时同步工资表”自动化：触发、循环、查找、分支、修改或新增，没有一步交给大模型。 04 ## 第一个月：问题不在功能，在稳定性 5 月 7 日第一版全流程跑通：早上 9 点按工地推任务，下午 5 点提醒工头确认，晚上 7 点二次提醒。5 月 8 日业务咨询师在内部群写下：“OpenClaw 的稳定性一直是最大的问题。这个不解决，将无法商用。”不稳定有四个来源，解决办法不神秘：锁定稳定版本、改动前先备份、加配置让重连更稳，再写一份客户能照着看的处理说明。同一时期我们也想清楚了输入问题：工头站在工地上，手上有灰，答案不在表格里，在输入法里，手机装一个语音输入法，说完看到字再发。这件事没有做任何开发。 第一个月的真正问题。四个来源，三个工程手段，加一份故障处理说明。 05 ## 5 月 27 日线下会：把“工地日志”说清楚 5 月 27 日我们去了客户公司，项目经理第一次到场。会上把工地日志七项内容逐条列出，由独立智能体每天固定时间生成一份 Daily Site Log；大包与非大包项目日志不同，在项目表上加一个属性字段；VO 由工头现场用自然语言提交，业主有两种签字习惯，所以要能导出两种 Excel 格式。这次会也把第一阶段目标收成一个：成本可视化。 5 月 27 日会议纪要里的工地日志七项框架。5 月 22 日工程师的工地日志智能体方案：进群但默认不说话，识别九类信息写入结构化表，每天生成一份日志。 06 ## 6 月：上线、升级、改规则 6 月 1 日到 3 日部署到老板办公室的 Mac mini，6 月 15 日全公司正式使用。两周内改了四件事：多维表格升到专业版（单表 2 万行）、时区从北京改为新西兰、工资表对群成员不可见并另建技术对接群、VO 工时改为在日工时明细里打标记。6 月 17 日一次人工核查抓出四类数据错误，系统一条都没拦住；同一天老板用一句自然语言改了一条规则：不说工时的信息记到工地日志或事件描述，只有可结算工时才进日工时明细表。 6 月 6 日的成本驾驭舱：15 个项目、6,013 小时、210 条记录。含金额的卡片已打码。6 月 17 日的四类错误。这也是老板安排专人每天盯数据的原因。6 月 17 日用一句话改规则，智能体确认后当场补齐两条记录，现有工时一条没动。项目名已打码。 07 ## 6 月 27 日一季度交付汇报 交付汇报的口径：系统完成约 85%，12 个核心表格，20 个项目入驻，334 条日常记录，累计 1 万多小时工时，当日仪表盘显示 12,149 小时。管理层从一屏看经营状态，每张表可以用自然语言问它。老板当场定了两件事：安排专人每天盯系统使用情况；把员工从 Excel 转到飞书系统，给一段缓冲期再逐步强制。审批环节没有进系统，成本审批仍在线下。二季度起客户团队自己操作，我们做教练。 三周之间的增长：项目、工时、记录、核心表四项都翻了一截。6 月 27 日汇报材料里的管理视图。右侧含人名的分布饼图已裁去。 08 ## 交付之后：退群、修复、移交 7 月 9 日我们的人退出了工时记录群和工地管理群，群交还客户，也是为了保护客户隐私。7 月 24 日客户自建的新智能体顶掉了原智能体的飞书绑定，工程师当天远程修复，并写了一份“如何创建多个智能体而不破坏原绑定关系”的操作指引，这是我们交接文档的缺口。7 月 29 日群里一位工人 @ 智能体查整周工时，它回复合计 49.5 小时并逐日列明细，没有人打开过 Excel。9 月 1 日确认整体运行稳定，进入常态化使用，群主移交客户。 交付之后的三步：退出、修复、移交。6 月 29 日整理的项目运营系统结构图。客户公司名已遮盖。 09 ## 三个月留下的四条教训 回头看，最贵的不是开发，是这四个判断。 稳定性是第一个功能：锁版本、先备份、写一份客户能照着做的故障处理说明，排在任何模型能力之前。 入口决定采用率：工头站在工地上怎么输入，决定系统有没有人用。下拉选择、语音转文字、工人不需要账号，没有一件是模型做的。 上线不是终点，数据要有人天天盯：6 月 17 日四类错误没有一条被系统拦住。老板安排专人每天看，是系统能进入常态化的真正原因。 交出去之前，把“怎么不弄坏它”写下来：角色权限、新建智能体的步骤、出问题先看哪里，要和系统一起交付。 10 ## 现在怎样了？ 系统由客户自己运营，我们的人已经不在工时统计群里。二季度客户团队在同一套飞书加 OpenClaw 的框架上自己开始做第二件事：给关联的设计公司做 H1 保温合规报告的自动计算，我们做教练。一套系统能不能留下来，取决于最不想改变的那个人愿不愿意用它。 三个月时间线，每个节点都能在项目记录里找到日期。 证据边界：本文所有数字来自项目群聊记录、6 月 27 日一季度交付汇报纪要与当日仪表盘截图，日期均已标注；“12 个核心表格、20 个项目、334 条记录、1 万多小时”是 6 月 27 日的截止快照，此后系统由客户运营，我们不再掌握最新数字。客户为新西兰南岛建筑施工公司，已匿名；截图中的公司名与人名已裁除或遮盖，含金额与项目名的区域已打码，工资数字未采用。 ## 一手参考 飞书多维表格（官方产品页） ↗Lark Base（官方产品页，英文） ↗ FAQ · 快速回答 ## 新西兰建筑公司的 AI 落地案例里，工地工时管理 AI 做哪一步，人做哪一步？ AI 做四件事：早上按工地推送任务、解析工头在群里说的话并写进日常记录和日工时明细、生成工地日志、汇总成本看板。人做三件事：工头每天确认实际工时，项目经理编辑任务和审批跨工地调拨，老板和会计核对时薪与成本。工资计算走多维表格自动化，不走大模型；审批不在系统里。 ## 需要换掉 Excel，或者让工人装新 App 吗？ 不需要装新 App。工人只在飞书或 Lark 的群里收通知、说一句话，不需要公司账号；不想打字的用语音转文字。Excel 记工时是逐步退出的：6 月 15 日系统正式使用后，老板给了员工一段缓冲期，再逐步强制在系统里记录。 ## 工地工时管理系统多久能上线？ 这个项目 4 月 8 日启动，5 月 7 日第一版工作流跑通，6 月 1 日部署到客户的 Mac mini，6 月 15 日全公司正式使用，约 10 周。其中开发本身不到一半时间，剩下的花在稳定性、飞书企业认证、表结构确认和现场输入习惯上。 ## 工资和成本数据谁能看到？ 分角色。工头只看本工地当日派工与工时，看不到财务成本和其他工地；项目经理看所有工地的工时汇总、进度、VO 与成本；会计通过独立的关联表看工时与成本的计算结果，时薪表和工时表分离；工资汇总表对工时统计群的普通成员不可见（6 月 16 日调整）。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # New Zealand AI Implementation Case Study: Site Timesheets and Job Costing for a Construction Company, Run from a Chat Group | BEE Sigma URL: https://beesigma.com/insight-case-nz-construction-site-timesheet-ai-en Summary: AI implementation case study, New Zealand (Lark Base, OpenClaw, Mac mini): a South Island construction company where hours were recalled by foremen after… New Zealand AI Implementation Case Study: Site Timesheets and Job Costing for a Construction Company, Run from a Chat Group | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · AI Implementation Case Study · NZ Construction ## New Zealand AI Implementation Case Study: Site Timesheets and Job Costing for a Construction Company, Run from a Chat Group Reviewed 14 Sep 20268 minBEE Sigma Delivery This New Zealand AI implementation case study is about a South Island construction company that works as both head contractor and carpentry subcontractor, with dozens of workers across close to twenty projects. Hours were recalled by foremen after work and totalled in Excel; the owner could not see labour cost by project. Kickoff was 8 April 2026, the first workflow ran on 7 May, the whole company went live on 15 June. At the first-quarter handover on 27 June: 12 core tables, 20 projects, 334 daily records and more than 10,000 hours in Lark Base. The diagrams below tell what happened in between. 01 ## How were hours recorded before? Internal communication ran in WeChat groups, external in email, and hours in Excel: a foreman recalled after work who had done how many hours, and the accountant totalled it at month end for payroll. On 9 April the owner set out the need: a system where site workers’ hours and projects are filled in daily and tracked, so that site managers, workers, accounting and admin can each quickly see what they need; on 28 April VO (Variation Order) records and a site diary were added. Taken apart, that is three things: who went where for how long, what happened on each site each day, and how much labour each project has consumed. No table connected them. Hours before and after. Memory and Excel on the left; one sentence in the group on the right. 02 ## Where to cut first: hours, not the whole system At the scoping session on 21 April we set phase one as hour tracking: the biggest data volume, happening every day, easiest to check for right and wrong; progress tracking went to phase two. Two principles were fixed that day: whatever a foreman fills in comes from dropdowns and simple checks, no free typing on a phone; voice is only used to call up the form, never to issue commands. On 22 April the plan became a five-step flow and four roles, with workers needing no company account, only an external Lark group for notices. The five-step flow from the 22 April plan: tasks in the morning, foreman confirms or edits, VOs on demand, actual hours before knock-off, system consolidates and generates the diary. (Chinese-language original.)The four roles in the same plan. Foremen cannot see financial cost; finance sees hours and cost through a separate private table, with pay rates kept apart from timesheets. (Chinese-language original.) 03 ## What did we connect? Still connect, don’t replace: workers still talk in the group, the owner still reads tables, and a layer of agents that write tables sits in between. One Lark group per site; the foreman posts completed work, incidents and photos and @-mentions the agent. The agents run on a Mac mini in the client’s office under OpenClaw, one each for hours, site diary and VOs. The foundation is one Lark Base: the daily hours detail table is the single source of hours and labour cost, and hours become payroll through a Base automation applying pay rates, with no LLM involved. The agent does not change rates, compute payslips or approve anything. From one sentence to the cost dashboard. Gold nodes are people; payroll never passes through an LLM.The table list before deployment on 1 June, plus the ‘Construction Overview | Cost Cockpit’ dashboard. (Chinese-language original.)The ‘sync hours to payroll’ automation built on 23 April: trigger, loop, lookup, branch, update or create. No step handed to a language model. 04 ## Month one: the problem was not features, it was stability On 7 May the first workflow ran end to end: tasks pushed by site at 9 a.m., confirmation prompts at 5 p.m., a second reminder at 7 p.m. On 8 May our business consultant wrote in the internal group: ‘OpenClaw’s stability has been the biggest problem. Until it is solved, this cannot go commercial.’ The instability had four sources, and the fix was not mysterious: pin a stable version, back up before changing anything, add configuration so reconnects hold, and leave a troubleshooting note the client can follow. In the same period we also settled the input question: a foreman stands on site with dust on his hands, so the answer is not in the tables but in the keyboard. A speech-to-text keyboard on the phone, see the words, then send. No code was written for this. The real problem of month one. Four sources, three engineering habits, plus a written troubleshooting note. 05 ## The on-site meeting on 27 May: defining the site diary On 27 May we went to the client’s office and the project manager attended for the first time. The meeting listed the seven items a site diary must record, handled by a separate agent that generates one Daily Site Log a day; lump-sum and non-lump-sum projects differ, so an attribute field went onto the project table; VOs are submitted by foremen in plain language, and because principals sign in two ways, two Excel formats had to be exportable. The meeting also narrowed phase one to one goal: cost visibility. The seven-item site-diary framework from the 27 May meeting notes. (Chinese-language original.)The engineer’s site-diary agent plan of 22 May: silent in the group, nine kinds of information into structured tables, one diary a day. (Chinese-language original.) 06 ## June: go-live, upgrade, rule changes From 1 to 3 June everything was deployed to the Mac mini in the owner’s office, and on 15 June the whole company went live. Four things changed within two weeks: the Base moved to the professional plan (20,000 rows per table), the time zone was corrected from Beijing to New Zealand, the payroll table was hidden from group members and a separate technical group created, and VO hours became a tag on the daily hours detail. A manual audit on 17 June caught four kinds of data error that the system had not stopped; the same day the owner changed a rule in one sentence: messages not about hours go to the site diary or event description, and only billable hours go to the daily hours detail. The cost cockpit on 6 June: 15 projects, 6,013 hours, 210 records. Cards with amounts are pixelated. (Chinese-language original.)The four error types of 17 June. This is why the owner assigned someone to watch the data daily.Changing a rule with one sentence on 17 June; the agent confirmed and completed two records on the spot without touching existing hours. Project names pixelated. (Chinese-language original.) 07 ## The first-quarter handover on 27 June The handover figures: system about 85% complete, 12 core tables, 20 projects onboarded, 334 daily records, more than 10,000 hours logged, with the dashboard that day showing 12,149. Management sees the business on one screen, and every table can be questioned in plain language. The owner decided two things in the room: assign one person to watch system usage every day, and move staff from Excel to Lark with a grace period before making it mandatory. Approvals stay outside the system; cost approval is still done offline. From the second quarter the client’s team operates and we coach. Three weeks of growth across projects, hours, records and core tables.A management view from the 27 June handover material. Pie charts containing names were cropped out. (Chinese-language original.) 08 ## After handover: exit, fix, hand over On 9 July our people left the hours group and the site management group and handed them back, which also protects the client’s privacy. On 24 July a new agent the client had built displaced the original agent’s Lark binding; the engineer fixed it remotely the same day and wrote a guide on how to add agents without breaking the existing binding, a gap in our handover documents. On 29 July a worker @-mentioned the agent for a whole week’s hours and got 49.5 hours with a day-by-day breakdown; nobody opened Excel. On 1 September the system was confirmed stable and in routine use, and group ownership was transferred to the client. Three steps after delivery: exit, fix, hand over.The project operations system diagram drawn up on 29 June. The client’s company name is masked. (Chinese-language original.) 09 ## Four lessons from three months Looking back, the expensive part was not development. It was these four judgements. Stability is the first feature: Pin versions, back up first, write a troubleshooting note the client can follow. These come before any model capability. The entry point decides adoption: How a foreman standing on site enters data decides whether the system is used. Dropdowns, speech-to-text, no accounts for workers: none of those was done by a model. Go-live is not the finish line; someone has to watch the data daily: None of the four error types on 17 June was caught by the system. The owner assigning one person to check daily is the real reason it reached routine use. Before handing over, write down how not to break it: Role permissions, the steps for creating a new agent, where to look first when something goes wrong: these ship with the system. 10 ## Where is it now? The system is run by the client, and our people are no longer in the hours group. In the second quarter the client’s team started a second project on the same Lark plus OpenClaw foundation: automating H1 insulation compliance calculations for their associated design practice, with us coaching. Whether a system stays depends on whether the person least willing to change is willing to use it. Three months on a timeline; every node has a date in the project record. Evidence boundary: every number in this article comes from the project group chat records, the notes of the first-quarter handover on 27 June and that day’s dashboard screenshots, each dated in the text. ‘12 core tables, 20 projects, 334 records, 10,000+ hours’ is a snapshot as of 27 June; the client has run the system since, and we no longer hold current figures. The client is a construction company on New Zealand’s South Island and is anonymised; company and personal names in screenshots have been cropped or masked, areas containing amounts and project names are pixelated, and payroll figures were not used. ## Primary sources Lark Base (official product page) ↗Feishu Base (official product page, Chinese) ↗ FAQ · Quick answers ## In this New Zealand construction AI implementation case study, what does the AI do for site hour tracking and what do people do? The AI does four things: pushes tasks by site each morning, parses what foremen say in the group and writes it into daily records and the daily hours detail, generates the site diary, and consolidates the cost dashboard. People do three: foremen confirm actual hours daily, the project manager edits tasks and approves cross-site transfers, and the owner and accountant check pay rates and cost. Payroll is calculated by a Base automation, not by a language model, and approvals are not in the system. ## Do we need to replace Excel, or make workers install a new app? No new app. Workers only receive notices and say one sentence in a Lark or Feishu group, with no company account needed; anyone who does not want to type uses speech-to-text. Excel timesheets were phased out gradually: after the system went live on 15 June, the owner gave staff a grace period before making recording in the system mandatory step by step. ## How long does a site hour tracking system take to go live? This project started on 8 April, the first workflow ran end to end on 7 May, it was deployed to the client’s Mac mini on 1 June and the whole company was using it on 15 June, about ten weeks. Development itself took less than half of that; the rest went on stability, Feishu enterprise verification, agreeing the table structure and on-site input habits. ## Who can see payroll and cost data? It depends on the role. Foremen see only their own site’s roster and hours for the day, not financial cost or other sites; the project manager sees hours, progress, VOs and cost across all sites; the accountant sees the calculated hours and cost through a separate linked table, with pay rates kept apart from timesheets; the payroll summary table is hidden from ordinary members of the hours group (changed on 16 June). DT BEE Sigma Delivery The front-line team that plugs workflows into the systems businesses already run — methods drawn from delivered projects. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 技术落地实践案例：一家建筑公司发票自动分流 91% 的三个月实录 | BEE Sigma URL: https://beesigma.com/insight-case-nz-construction-invoice-automation Summary: 新西兰 AI 技术落地实践案例（建筑业）：供应商发票每月 500 到 1000 张，一位工料测量师手工分类、匹配采购单、提交审批。三个月后，一个完整会计月里 798 份发票文档中 730 份由 AI 自动识别分流，人工复核降到 68 份。 新西兰 AI 技术落地实践案例：一家建筑公司发票自动分流 91% 的三个月实录 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · AI 落地案例 · 新西兰建筑 ## 新西兰 AI 技术落地实践案例：一家建筑公司发票自动分流 91% 的三个月实录 内容复核 2026.09.118 分钟BEE Sigma 落地团队 这篇新西兰 AI 落地案例的主角是一家建筑总承包商，项目横跨住宅、商业地产和酒店翻新，供应商发票每月 500 到 1000 张，接收、分类、匹配采购单、提交审批、跟踪逾期全靠一位工料测量师（QS）手工完成。2026 年 6 月我们接入，7 月用真实发票试运行，8 月全量校准。按客户会计月（7 月 26 日到 8 月 25 日）统计：798 份发票相关文档，730 份由 AI 自动分流，人工复核 68 份，自动分流比例 91.48%（2026-09-01 项目复盘口径，未经客户书面确认）。下面用图讲中间三个月发生了什么。 01 ## 原来每个月初，发票是怎么处理的？ 客户的系统很典型：发票进 Outlook，文件存 OneDrive，审批走 ApprovalMax，记账在 Xero。每月 26 日到次月 3 日是跨月窗口，月初两三天可能压着 600 到 700 张，QS 要逐张判断工地、项目类型、采购单和审批人。审批人和工地的对应关系不在任何文档里，在几个人的经验里。 合作前后一张发票的走法对照。左边五步全在一位 QS 手上；右边 Agent 接、人只批。 02 ## 我们接进了什么？ 原则是连接而不是替换：一个系统都没换。Agent 跑在客户办公室的一台 Mac mini 上，用 OpenClaw 调度，通过只读邮件权限监听专用发票邮箱，识别、归档、匹配 PO、提交审批、每小时回写状态。人留在三个地方：QS 处理人工队列，项目经理在 ApprovalMax 里审批，财务在 Xero 里付款和锁账。AI 不批、不付、不改账。 发票的旅程。金色节点是人必须出手的两处：审批与付款。 03 ## 第一个月：问题不在识别率，在规则 7 月 21 日第一次量化：407 张候选发票只有 63 张进入审批，完成率 15.5%。拆开看，真正识别失败只有 12 张，不到 3%；卡住的是 173 张规则不完善和 132 张信息缺失。模型读发票读得很好，读完不知道该交给谁。7 月 27 日我们把 111 张问题发票归成 7 类，请客户逐类书面确认处理方式，工作流恢复自动运行。 7 月 21 日的阻塞分布。规则和映射才是工作量，识别失败不到 3%。7 月 25 日团队内部的分流设计图：当轮 151 张里 112 张仍需人工介入，把原因分成规则缺失、信息可补全、信息不可获得三类，只有第三类才该长期留给人。 04 ## 发票到底有多少张？ 8 月初，发票总数这个最朴素的数字对不上了：7 月记录从 369 张变成 460、815，最后确认标准发票 517 张。根因是识别、拆分、去重混在一条流水线里，加上一次“清理已处理的 6 月发票”指令被 Agent 按发票日期而非接收日期执行，误清了部分 7 月发票（有备份，当天找回）。改法是把总账做成逐层可对账的审计链，仪表盘从此分成“上一周期正式结果”和“当前周期进行中”两块。 8 月 10 日仪表盘里 7 月会计周期的正式结果：收到 674 份，等待信息 529 份，完成率只有 4%。这是真实起点，我们把它留在了仪表盘里。8 月 6 日改版后的仪表盘：上半部是进行中的 8 月周期，下半部是已关闭的 7 月周期，数字来源分开写清楚。 05 ## 8 月 4 日的转向：先保证不断流 8 月初我们意识到方向偏了：工程师一直在优化识别规则让更多发票自动通过，而客户要的是业务先不断流。AI 判断不了的发票，哪怕信息不全，也要有序分到对应的人手里，而不是堆进一个 Manual Review 黑箱。这是整个项目最重要的一次转向，之后的设计都围绕“剩下的那部分怎么交给人”。 8 月 4 日之后人工队列的六项设计。人工队列不是残余，是设计对象。8 月 7 日的处理漏斗：676 份进入，650 份进入审批（96.87%）。漏斗只回答“自动化了多少”，不回答“剩下的去了哪”。 06 ## 8 月：全量校准 8 月下旬收口，人工复核队列从 300 多份降到 181、110、81。8 月 30 日交付口径：自然月 775 份，694 份自动分流，89.55%。9 月 1 日按客户会计周期（7 月 26 日到 8 月 25 日）重算：798 份，730 份自动分流，人工复核 68 份，91.48%。两个口径都写在这里，因为客户看会计月，我们内部看自然月。 自动分流比例从 4% 到 91.48%，每个数字都标了日期和口径。8 月 27 日仪表盘（7 月周期视图）：收到 442、已推送 ApprovalMax 314、处理完成率 92%、审批完成率 78%。9 月 3 日切到 8 月周期：待审批 204、审批逾期 7、QS 待处理 60、加急 30。 07 ## 谁能看金额，谁不能？ 仪表盘做了两版：Admin 版给老板和财务，含金额；审批人版只看自己名下待办，不含金额。多维表格里建了 Approver、Admin、AI Master 三个角色，Agent 用 AI Master 身份写数据，看不到不该看的。 8 月 31 日按角色隔离后的权限配置：审批人版仪表盘对 Admin 只读，审批人对 Admin 版无权限。 08 ## 平台的边界，AI 不越 9 月 1 日客户需要把一批 7 月发票的日期改到 8 月。ApprovalMax 允许改日期，但改完审批流程重置，而 API 没有 approve 接口。我们对 37 份执行：19 份成功，9 份权限冲突跳过，9 份已 rejected 不再动。重置后的审批由有权限的人手动点。这条后来写进交付规则：凡是审批、付款、对外承诺，AI 准备、AI 核对，人最后一按。 平台不允许程序替人批，那就让人批。不绕。 09 ## 三个月留下的四条教训 回头看，最贵的不是技术，是这四个判断。 只看识别率会走错方向：识别失败不到 3%，完成率却只有 15.5%。真正的工作量在规则和映射，问不出来的要设计成人工路径。 总账要有独立的审计链：“到底收到多少张”答不上来，信任会立刻回到零。识别、去重、统计分层，每层能对账。 AI 按字面执行指令：涉及删除的指令要先试跑、有备份、有第二个数据源交叉。 把过程讲给客户听：从“功能做完”转向“业务跑通”，少用开放式提问，多用固定步骤和选择题，主动发现并闭环问题。 10 ## 现在怎样了？ 8 月 30 日技术阶段交付，9 月进入优化期：修审批提醒重复、供应商归类误判、锁账后的日期调整。最好的验收不是数字：9 月初客户把我们介绍给了他们合作的建筑保险经纪，另一个“文档重、合规重”的行业。 三个月时间线，每个节点都能在项目记录里找到日期。 证据边界：本文所有数字来自项目群聊记录与 9 月 2 日项目复盘文档，日期均已标注；91.48% 按客户会计周期统计，89.55% 按自然月统计，两者分母都包含非发票类文件。客户为新西兰建筑总承包商，已匿名；截图中的公司名与人名已裁除或遮盖，含金额的仪表盘未采用。 ## 一手参考 ApprovalMax 公开 API 说明（ApprovalMax 官方博客） ↗ FAQ · 快速回答 ## 新西兰建筑公司的 AI 落地案例里，发票自动化 AI 做哪一步，人做哪一步？ AI 做接收、识别、归档、匹配采购单、提交审批和状态回写；人做三件事：处理 AI 判断不了的发票（Manual Review）、在 ApprovalMax 里审批、在 Xero 里付款和锁账。涉及批准和付款的动作，AI 不碰。 ## 需要换掉 Xero 或 ApprovalMax 吗？ 不需要。整个项目一个系统都没换，Agent 通过 Outlook、OneDrive、ApprovalMax 和 Xero 之间的接口工作，运行在客户自己的 Mac mini 上，发票原件不离开客户环境。 ## 自动分流 91% 意味着什么？剩下 9% 怎么办？ 意味着一个会计月 798 份文档里 730 份不需要人看第一眼。剩下 68 份进入按原因分类的人工队列，QS 补齐信息后 Agent 自动接着走。剩下的这部分才是设计重点，不是残余。 ## 发票自动化多久能看到效果？ 这个项目 6 月接入，7 月试运行，8 月全量校准，第一个完整会计月的数字在 9 月初出来。我们有一条销售对账工作流 11 天上线的记录，跟单、对账类流程一般按 2 到 4 周规划试点；发票这类涉及多方映射规则的，按三个月准备。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # New Zealand AI Implementation Case Study: 91% Invoice Automation at a Construction Company | BEE Sigma URL: https://beesigma.com/insight-case-nz-construction-invoice-automation-en Summary: AI implementation case study, New Zealand (Outlook, OneDrive, ApprovalMax, Xero): five hundred to a thousand supplier invoices a month, one quantity… New Zealand AI Implementation Case Study: 91% Invoice Automation at a Construction Company | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · AI Implementation Case Study · NZ Construction ## New Zealand AI Implementation Case Study: 91% Invoice Automation at a Construction Company Reviewed 11 Sep 20268 minBEE Sigma Delivery A New Zealand construction head contractor with residential, commercial and hotel-refurbishment projects receives 500 to 1,000 supplier invoices a month. Receiving, sorting by site, matching purchase orders, submitting approvals and chasing overdue items all sat with one quantity surveyor (QS). We connected in June 2026, ran real invoices in July and calibrated at full volume in August. On the client’s accounting month (26 July to 25 August): 798 invoice-related documents, 730 routed automatically, 68 in manual review, an automatic routing rate of 91.48% (project review of 1 September 2026, not confirmed in writing by the client). The diagrams below tell what happened in between. 01 ## How were invoices handled at the start of each month? The client’s stack is typical: invoices arrive in Outlook, files live in OneDrive, approvals run through ApprovalMax, the books are in Xero. The 26th to the 3rd is the cross-month window, when 600 to 700 invoices could pile up and the QS had to judge each one’s site, project type, purchase order and approver. The mapping from approver to site was not written down anywhere. It lived in a few people’s experience. One invoice, before and after. Five steps on one QS’s desk on the left; the agent receives and people only approve on the right. 02 ## What did we connect? The principle was connect, don’t replace: not one system was swapped. The agent runs on a Mac mini in the client’s office, orchestrated by OpenClaw, watches a dedicated mailbox through read-only permission, recognises, files, matches POs, submits approvals and writes status back hourly. People stay in three places: the QS handles the manual queue, project managers approve in ApprovalMax, finance pays and locks the month in Xero. The AI does not approve, pay or post. An invoice’s journey. The gold nodes are the two places where a person must act: approval and payment. 03 ## Month one: the problem was rules, not recognition The first count on 21 July: 407 candidate invoices, 63 into approval, 15.5% complete. Broken down, only 12 were genuine recognition failures, under 3%; what blocked the rest was 173 incomplete rules and 132 missing information. The model read invoices very well and then did not know who to hand them to. On 27 July we grouped 111 problem invoices into 7 classes and asked the client to confirm in writing how each should be handled. The workflow went back to running automatically. The blockers on 21 July. Rules and mappings were the work; recognition failures were under 3%.The team’s internal routing design from 25 July: 112 of 151 invoices still needed a person, split into missing rules, information that can be completed, and information that cannot be obtained. Only the third category should stay with people long term. (Diagram labels are in Chinese.) 04 ## How many invoices are there, exactly? In early August the simplest number stopped reconciling: July’s count went from 369 to 460 to 815 and finally to 517 confirmed invoices. Root causes: recognition, splitting and de-duplication were mixed into one pipeline, and an instruction to clean up processed June invoices was executed by the agent on invoice date rather than received date, removing some July invoices (backed up, recovered the same day). The fix was an independent audit chain that reconciles layer by layer, and a dashboard split into a closed cycle and a live cycle. The July cycle on the 10 August dashboard: 674 received, 529 awaiting information, 4% complete. The real starting point, and we left it there.The dashboard after the 6 August redesign: live cycle above, closed cycle below, data sources stated separately. 05 ## The turn on 4 August: keep the flow moving first In early August we realised our direction had drifted. The engineer kept tuning recognition rules so more invoices would pass automatically, while the client needed the business to keep flowing first. Invoices the AI could not decide on, even with incomplete information, had to be routed to the right person in an orderly way rather than piled into one Manual Review black box. That was the most important turn in the project, and everything after it was designed around how the remainder gets handed to people. The six design elements of the manual queue after 4 August. The manual queue is a design object, not a leftover.The funnel on 7 August: 676 in, 650 into approval (96.87%). A funnel answers how much was automated, not where the rest went. 06 ## August: full-volume calibration Late August was close-out; the manual queue fell from 300-plus to 181, 110, 81. Handover figure on 30 August: 775 documents in the calendar month, 694 routed automatically, 89.55%. Recalculated on 1 September for the client’s accounting period (26 July to 25 August): 798 documents, 730 routed, 68 in manual review, 91.48%. Both bases are stated because the client looks at accounting months and we looked at calendar months. From 4% to 91.48%, each figure with its date and basis.27 August (July cycle view): 442 received, 314 pushed to ApprovalMax, 92% processing completion, 78% approval completion.3 September, August cycle: 204 pending approval, 7 overdue, 60 for the QS, 30 urgent. 07 ## Who can see the amounts, and who cannot? Two dashboards: the Admin version for owners and finance shows amounts; the Approver version shows each approver only their own queue, without amounts. Three custom roles in Lark Base, Approver, Admin and AI Master, and the agent writes under AI Master, unable to see what it should not. Role-based isolation as configured on 31 August: the Approver dashboard is read-only for Admin, and approvers have no access to the Admin dashboard. 08 ## The platform’s boundary is where the AI stops On 1 September the client needed a batch of July invoice dates moved into August. ApprovalMax allows the change, but it resets the approval workflow, and the API has no approve endpoint. Of 37 invoices: 19 updated, 9 skipped on a permission conflict, 9 left alone because they were already rejected. The reset approvals were clicked by people with permission. It became a delivery rule: for approval, payment or external commitments, the AI prepares and checks, and a person presses the last button. If the platform will not let a program approve on a person’s behalf, a person approves. No workaround. 09 ## Four lessons from three months Looking back, the expensive part was not the technology. It was these four judgments. A recognition rate alone points the wrong way: Failures under 3%, completion 15.5%. The real work is rules and mappings; what cannot be asked out must be designed as a manual path. The ledger needs an independent audit chain: When you cannot say how many invoices arrived, trust goes straight back to zero. Recognition, de-duplication and counting are separate layers that reconcile. AI executes instructions literally: Anything that deletes needs a dry run, a backup and a second data source. Tell the client about the process: Move from feature done to business running; fixed steps and multiple choice instead of open questions; find and close problems proactively. 10 ## Where is it now? The technical phase was handed over on 30 August and September is an optimisation period: duplicate reminders, supplier misclassification, post-lock date adjustments. The best acceptance test was not a number: in early September the client introduced us to their construction insurance broker, another industry heavy on documents and compliance. Three months on a timeline; every node has a date in the project record. Evidence boundary: every figure in this article comes from the project chat record and the project review document of 2 September, each dated. 91.48% is measured on the client’s accounting period and 89.55% on the calendar month; both denominators include non-invoice files. The client is a New Zealand construction head contractor and is anonymised; company and personal names in screenshots have been cropped or masked, and dashboards showing amounts were not used. ## Primary sources What you can do with the ApprovalMax public API (ApprovalMax blog) ↗ FAQ · Quick answers ## In this New Zealand AI implementation case study, what does the AI do and what do people do? The AI receives, recognises, files, matches purchase orders, submits approvals and writes status back. People do three things: handle the invoices the AI cannot decide on (Manual Review), approve in ApprovalMax, and pay and lock the month in Xero. Anything that approves or pays, the AI does not touch. ## Do we need to replace Xero or ApprovalMax? No. Not a single system was replaced in this project. The agent works through the interfaces between Outlook, OneDrive, ApprovalMax and Xero, runs on the client’s own Mac mini, and original invoices never leave the client’s environment. ## What does a 91% automatic routing rate mean, and what about the other 9%? It means that in one accounting month, 730 of 798 documents needed no first look from a person. The remaining 68 went into a manual queue sorted by cause; once the QS fills in the missing information, the agent picks them up again. That remaining part is the design focus, not a leftover. ## How long until invoice automation shows results? This project connected in June, ran real invoices in July and calibrated at full volume in August; the first full accounting month’s numbers came out in early September. One sales reconciliation workflow in our records went live in 11 days; follow-up and reconciliation workflows are usually planned as a two-to-four-week pilot. Invoices, with their many mapping rules across parties, should be planned as a three-month effort. DT BEE Sigma Delivery The front-line team that plugs workflows into the systems businesses already run — methods drawn from delivered projects. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰企业管理软件比较（2026）：财务、CRM、ERP 怎么选，以及为什么多数企业不必换系统 | BEE Sigma URL: https://beesigma.com/insight-nz-business-software-comparison Summary: Xero 还是 MYOB？HubSpot 还是 Pipedrive？要不要上 ERP？这篇把新西兰中小企业常用的管理软件按财务、CRM、ERP 与库存、协作四类做一次实用比较，给出选型判断标准，并解释一个反直觉的结论：多数企业的问题不在软件，而在软件之间没人串——这正是 AI Agent 该接的活。 新西兰企业管理软件比较（2026）：财务、CRM、ERP 怎么选，以及为什么多数企业不必换系统 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 新西兰 · 选型指南 ## 新西兰企业管理软件比较（2026）：财务、CRM、ERP 怎么选，以及为什么多数企业不必换系统 内容复核 2026.09.0610 分钟BEE Sigma 研究组 01 ## 先说结论：比较软件之前，先比较你的流程 在新西兰做企业管理软件比较，最容易掉进的坑是拿功能清单对功能清单。Xero 和 MYOB 谁的报表多、HubSpot 和 Pipedrive 谁的管道好看，这些比较网上到处都是，但很少回答老板真正的问题：换了之后，团队的活会不会少一点，数据会不会终于对得上。 我们服务新西兰与中国两地企业时反复看到同一件事：软件本身大多够用，问题出在软件之间。询价在邮箱、报价在 Word、订单在微信群、发票在 Xero、客户资料在某个人的脑子里，每一步之间都靠人抄。所以这篇比较分四类讲清各家的定位，最后回到那个反直觉的结论：多数企业不必换系统，需要的是把系统串起来的那一层。 02 ## 财务与会计：Xero、MYOB、Reckon 这是新西兰中小企业渗透率最高的一类软件，三家都能覆盖开票、银行对账、GST 申报和工资的基本面，差别在生态与习惯。 Xero惠灵顿起家的云会计软件，新西兰事实上的默认选项。优势是银行直连与第三方应用生态庞大，绝大多数会计师事务所熟悉它；库存、项目管理等进阶需求通常要靠外接应用补齐。 MYOB澳新两地的老牌选手，从小企业版到面向中型企业的 MYOB Acumatica（原 MYOB Advanced）覆盖面更宽。如果企业在澳洲也有实体，或未来要长到需要 ERP 的规模，MYOB 的向上迁移路径更连贯。 Reckon同样来自澳洲，定位偏实惠，适合流程简单、对生态要求不高的小团队。选它的企业通常更在意成本而不是集成。 怎么选问会计师用什么、问银行支持什么、问你要接的其他软件支持什么。三问之后答案基本就出来了。会计软件是财务的“账本”，账本尽量不要频繁更换。 03 ## CRM：HubSpot、Pipedrive、Zoho、Salesforce CRM 是新西兰企业买得最多、也最容易闲置的一类软件。原因不在产品，在于它要求销售把发生在电话、邮件和微信里的事再手动录一遍。比较之前先接受这个事实，再看定位。 HubSpot免费版门槛低、营销与销售一体，适合以内容和线上获客为主的团队；规模上去后按模块付费，成本曲线要提前算。 Pipedrive以销售管道为中心，界面直观，小团队上手最快；营销自动化与客服模块相对薄。 Zoho CRM性价比高、模块齐全，与 Zoho Books、Zoho Desk 同生态，适合预算有限但想要一站式的中小企业；新西兰本地实施服务商相对少。 Salesforce功能与可定制性的天花板，实施与维护成本同样是天花板，更适合有专职管理员的中大型销售组织。 怎么选先回答一个问题：谁来录数据？如果答案还是“销售自己”，再贵的 CRM 也会空着。我们在《最好的 CRM，是销售团队从来不需要打开的 CRM》里写过解法：让 Agent 从对话里把要素抓出来、确认后写进 CRM，人只做确认。 04 ## ERP 与库存：什么时候才真的需要 “要不要上 ERP”是新西兰做贸易、制造和多渠道零售的老板常问的问题。判断标准很简单：当库存、采购、订单和财务这四件事已经在四个地方各记一份、月底对不上时，才需要一个统一的数据底座。在此之前上 ERP，往往是给自己找了一个更贵的 Excel。 Cin7 / Unleashed两家都是奥克兰起家的库存与订单管理软件，与 Xero 打通顺畅，适合有实体商品、多渠道销售（网店、批发、门店）的中小企业。它们不是完整 ERP，但常常已经够用。 MYOB Acumatica面向中型企业的云 ERP，财务、分销、制造、项目模块完整，澳新本地实施伙伴多，从 MYOB 小企业版长上来最顺。 NetSuiteOracle 旗下的全球云 ERP，多法人、多币种、多国合规能力强，适合已经跨国经营或计划快速扩张的企业，实施周期与预算都要按季度算。 Odoo开源模块化，从 CRM、库存到会计可以按需装，自由度高。前提是有自己的技术团队或可靠的实施方长期维护，否则自由度会变成负担。 05 ## 协作与沟通：Teams、Slack、微信、飞书 这一类经常被排除在“管理软件”之外，但它恰恰是大多数业务真正发生的地方。新西兰本地企业以 Microsoft 365 加 Teams 为主，科技与创意团队偏 Slack；华人经营的企业则几乎必然同时用微信或企业微信对客户、用飞书或钉钉管内部，还要和本地伙伴在 Teams 上开会。 比较协作工具的意义不大，因为你很难让客户换掉他的微信。真正该问的是：这些对话里的订单、承诺和付款截图，怎么不靠人抄就进到 CRM、库存和 Xero 里去。这就是下一节的主题。 06 ## 比较之后：为什么多数企业不必换系统 把四类软件比完，回到开头的结论。新西兰中小企业真正缺的通常不是一款更好的软件，而是软件之间那个“串起来”的人，而这个人今天可以是 AI Agent。三个症状对照一下，中两条以上，问题就不在选型。 症状一：同一份数据在三个地方各记一遍订单在微信群、出库在 Unleashed、发票在 Xero，月底靠人核对。这是集成问题，不是软件问题。 症状二：CRM 字段常年空着不是 CRM 不好用，是没人愿意二次录入。让 Agent 从邮件和对话里抓要素、人确认后写入，字段才会活。 症状三：换过系统，半年后回到老样子因为流程和习惯没变，只是换了个地方堆数据。先把流程跑顺，再谈要不要换。 我们的做法连接而不是替换：12 个职能 Agent 接进你已有的 Xero、CRM、库存系统与 Teams、微信、飞书，接管抄录、催办、核对这些重复动作，报价、成交、付款等决策留在人手里。新西兰企业流程自动化解决方案一页有四条工作流的具体做法。 07 ## 一张选型对照表 最后把判断标准压成一张表，比较任何一款新西兰企业管理软件时都可以照着过一遍。 会计师与银行你的会计师熟不熟、银行能不能直连。会计软件优先随会计师。 团队规模与录入者10 人以下的销售团队先看 Pipedrive 或 HubSpot 免费版；无论选谁，先定谁来录数据。 实物与多渠道有库存、有多渠道销售再考虑 Cin7 或 Unleashed；四件事在四处各记一份且对不上时，才考虑 ERP。 跨境与多币种同时在中国或澳洲经营、多法人多币种，看 MYOB Acumatica 或 NetSuite；单一新西兰实体大多不需要。 现有数据能不能留优先选能接入现有系统的方案。任何以“先换掉你现在的系统”开头的方案，都要格外警惕。 ## 一手参考 Xero（新西兰） ↗MYOB（新西兰） ↗新西兰国家 AI 战略与企业指引（MBIE） ↗ FAQ · 快速回答 ## 新西兰企业管理软件比较，中小企业到底选哪个？ 没有对所有企业最好的那一款，只有对你的流程最合适的组合。多数新西兰中小企业的合理起点是：会计跟着会计师选（大概率是 Xero 或 MYOB），CRM 从 Pipedrive 或 HubSpot 免费版起步，有实物库存再加 Cin7 或 Unleashed，四件事对不上账时才考虑 ERP。比软件更重要的是谁来录数据、系统之间怎么串，这正是 AI Agent 该接的活。 ## 新西兰最常用的会计软件是什么？Xero 和 MYOB 怎么选？ Xero 是新西兰事实上的默认选项，会计师普遍熟悉、银行直连和第三方生态最大；MYOB 在澳新两地都有深厚基础，从小企业版到中型企业的 MYOB Acumatica 迁移路径更连贯。选法：问你的会计师用什么、你的银行支持什么、你要接的其他软件支持什么。 ## 新西兰中小企业需要上 ERP 吗？ 多数不需要。只有当库存、采购、订单和财务已经在四个地方各记一份、月底对不上时，统一的数据底座才有价值。在那之前，Xero 加一款库存软件（Cin7 或 Unleashed）通常已经够用；真到需要时再看 MYOB Acumatica、NetSuite 或 Odoo。 ## 华人企业在新西兰用飞书、企业微信可以吗？怎么和 Teams、Xero 打通？ 可以，而且很常见：对客户用微信或企业微信，内部用飞书或钉钉，和本地伙伴在 Teams 开会。打通的正确方式不是让谁换工具，而是让 Agent 接进这些对话，把订单、承诺、付款截图抓成结构化数据，确认后写进 CRM、库存和 Xero。中新跨境物流客户的微信群订单闭环就是这样交付的。 ## 软件都选好了，AI Agent 在哪一层？ 在软件之间那一层。会计、CRM、库存、协作工具各自管好自己的数据，Agent 负责跨系统的抄录、催办、核对和提醒，人只做确认与决策。这也是为什么我们说连接而不是替换：不动你的系统，只补上串起来的那一层。想看你的企业该从哪条流程开始，可以先做免费的 AIM 落地评估。 RS BEE Sigma 研究组 梳理新西兰与中国两地的 AI 落地市场、供应商与政策，用可核对的事实说话。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Business Software in New Zealand Compared (2026): Accounting, CRM and ERP, and Why Most Companies Shouldn't Switch | BEE Sigma URL: https://beesigma.com/insight-nz-business-software-comparison-en Summary: Xero or MYOB? HubSpot or Pipedrive? Do you need an ERP at all? Business Software in New Zealand Compared (2026): Accounting, CRM and ERP, and Why Most Companies Shouldn't Switch | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · New Zealand · Buyer's Guide ## Business Software in New Zealand Compared (2026): Accounting, CRM and ERP, and Why Most Companies Shouldn't Switch Reviewed 6 Sep 202610 minBEE Sigma Research 01 ## The Conclusion First: Compare Your Process Before You Compare Software The easiest trap in a New Zealand business-software comparison is matching feature lists against feature lists. Whether Xero or MYOB has more reports, whether HubSpot or Pipedrive has the prettier pipeline — those comparisons are everywhere, and they rarely answer the owner's real question: after the switch, will the team do less manual work, and will the numbers finally reconcile? Working with businesses across New Zealand and China, we see the same thing again and again: the software is mostly fine; the problem sits between the systems. Enquiries live in the inbox, quotes in Word, orders in a WeChat group, invoices in Xero, and the client history in one person's head — with a human copying data at every hand-off. So this comparison covers four categories, then returns to the counter-intuitive conclusion: most companies don't need to switch systems. They need the layer that connects them. 02 ## Accounting: Xero, MYOB, Reckon This is the most widely adopted category among New Zealand SMEs. All three cover invoicing, bank reconciliation, GST returns and payroll; the differences are ecosystem and habit. XeroCloud accounting born in Wellington and the de facto default in New Zealand. Its strengths are bank feeds and a huge third-party app ecosystem, and almost every accounting practice knows it. Advanced needs such as inventory or job management usually rely on add-on apps. MYOBA long-standing trans-Tasman player with a wider span, from the small-business edition up to MYOB Acumatica (formerly MYOB Advanced) for mid-sized companies. If you also operate in Australia, or expect to grow into ERP territory, MYOB offers a more continuous upgrade path. ReckonAlso Australian, positioned on value, and suited to small teams with simple processes and little need for integrations. Companies that choose it usually care more about cost than connectivity. How to chooseAsk what your accountant uses, what your bank supports, and what the other software you want to connect supports. Three questions usually settle it. Accounting software is the ledger — don't change the ledger often. 03 ## CRM: HubSpot, Pipedrive, Zoho, Salesforce CRM is the category New Zealand businesses buy most and abandon most. The cause isn't the product: a CRM asks salespeople to re-type what already happened on the phone, in email and in WeChat. Accept that fact first, then look at positioning. HubSpotA low-friction free tier and marketing and sales in one place, good for teams that acquire through content and online channels. Costs step up by module as you scale, so model the curve early. PipedriveBuilt around the sales pipeline, intuitive, and the fastest for a small team to adopt; marketing automation and service modules are comparatively thin. Zoho CRMStrong value and a full module set within the Zoho ecosystem (Books, Desk and more), suited to budget-conscious SMEs that want one suite; local New Zealand implementation partners are fewer. SalesforceThe ceiling for capability and customisation, and also the ceiling for implementation and maintenance cost. It fits mid-to-large sales organisations with a dedicated administrator. How to chooseAnswer one question first: who enters the data? If the answer is still "the salespeople themselves", the most expensive CRM will sit empty. We wrote the fix in The Best CRM Is the One Your Sales Team Never Opens: let agents extract the facts from conversations and write them into the CRM after a human confirms. 04 ## ERP and Inventory: When You Actually Need One "Do we need an ERP?" is a standard question from New Zealand owners in trade, manufacturing and multi-channel retail. The test is simple: when inventory, purchasing, orders and finance are already recorded in four separate places and don't reconcile at month end, you need one data foundation. Before that point, an ERP is often just a more expensive spreadsheet. Cin7 / UnleashedBoth inventory and order management tools founded in Auckland, both integrate cleanly with Xero, and both suit SMEs with physical products and multiple channels (online, wholesale, retail). They are not full ERPs, but they are frequently enough. MYOB AcumaticaCloud ERP for mid-sized companies with complete finance, distribution, manufacturing and project modules, a deep bench of local implementation partners, and the smoothest path up from MYOB's small-business edition. NetSuiteOracle's global cloud ERP, strong on multi-entity, multi-currency and multi-country compliance. It fits companies already operating across borders or planning rapid expansion; budget the implementation in quarters. OdooOpen-source and modular — CRM, inventory and accounting can be installed as needed, with a lot of freedom. The precondition is an in-house technical team or a reliable implementation partner for the long run; otherwise freedom becomes a burden. 05 ## Collaboration: Teams, Slack, WeChat, Lark This category is often left out of "business software", yet it is where most of the business actually happens. Local New Zealand companies run mostly on Microsoft 365 and Teams, technology and creative teams lean to Slack, and Chinese-run businesses almost always use WeChat or WeCom with customers, Lark or DingTalk internally, and still meet local partners on Teams. Comparing collaboration tools has limited value, because you will struggle to make a customer give up WeChat. The real question is how the orders, commitments and payment screenshots inside those conversations reach the CRM, the inventory system and Xero without a human retyping them. That is the next section. 06 ## After the Comparison: Why Most Companies Shouldn't Switch With four categories compared, back to the opening conclusion. What New Zealand SMEs usually lack is not a better piece of software but the person who connects the pieces — and today that person can be an AI agent. Check three symptoms; if two apply, the problem isn't software selection. Symptom one: the same data recorded in three placesOrders in a WeChat group, dispatch in Unleashed, invoices in Xero, reconciled by hand at month end. That is an integration problem, not a software problem. Symptom two: CRM fields permanently emptyNot because the CRM is bad, but because nobody wants to enter data twice. Fields come alive when agents pull the facts from email and chat and a human confirms before writing. Symptom three: you switched systems and six months later nothing changedBecause the process and habits stayed the same; you only moved where the data piles up. Fix the flow first, then decide whether to switch. Our approachConnect, don't replace: 12 role agents plug into the Xero, CRM, inventory system, Teams, WeChat and Lark you already have, and take over transcribing, chasing and matching, while pricing, deals and payments stay human decisions. The workflow automation solutions page shows how across four workflows. 07 ## A Selection Checklist Finally, the criteria compressed into one checklist to run against any business software you evaluate in New Zealand. Accountant and bankDoes your accountant know it, and does your bank feed into it? Accounting software follows the accountant. Team size and who enters dataSales teams under ten should look at Pipedrive or HubSpot's free tier first; whichever you pick, decide who enters the data before you sign. Physical goods and channelsConsider Cin7 or Unleashed only once you hold inventory across channels; consider an ERP only when four systems record the same facts and disagree. Cross-border and multi-currencyOperating in China or Australia too, with multiple entities and currencies: look at MYOB Acumatica or NetSuite. A single New Zealand entity rarely needs either. Can existing data stayPrefer solutions that plug into what you already run. Any proposal that opens with "first replace your current system" deserves extra caution. ## Primary sources Xero New Zealand ↗MYOB New Zealand ↗New Zealand AI Strategy & business guidance (MBIE) ↗ FAQ · Quick answers ## Business software in New Zealand compared — which should an SME actually choose? There is no single best product, only the combination that fits your process. A sensible starting point for most New Zealand SMEs: accounting follows your accountant (most likely Xero or MYOB), CRM starts with Pipedrive or HubSpot's free tier, add Cin7 or Unleashed once you hold physical inventory, and consider an ERP only when four systems stop reconciling. More important than any product is who enters the data and how the systems connect — the job AI agents should take. ## What is the most common accounting software in New Zealand, and how do I choose between Xero and MYOB? Xero is the de facto default in New Zealand, with broad accountant familiarity, bank feeds and the largest third-party ecosystem. MYOB has deep roots on both sides of the Tasman and a more continuous path from its small-business edition to MYOB Acumatica for mid-sized companies. To decide: ask what your accountant uses, what your bank supports, and what the other software you want to connect supports. ## Do New Zealand SMEs need an ERP? Most don't. A unified data foundation earns its cost only when inventory, purchasing, orders and finance are recorded in four separate places and fail to reconcile at month end. Until then, Xero plus an inventory tool such as Cin7 or Unleashed is usually enough; when the time comes, look at MYOB Acumatica, NetSuite or Odoo. ## Can a Chinese-run business in New Zealand use Lark or WeCom, and how do they connect with Teams and Xero? Yes, and it is common: WeChat or WeCom with customers, Lark or DingTalk internally, Teams for local partners. The right way to connect them is not to make anyone switch tools but to let agents join those conversations, turn orders, commitments and payment screenshots into structured data, and write it into the CRM, inventory and Xero after confirmation. A China–NZ cross-border logistics client's WeChat-group order loop was delivered exactly this way. ## Once the software is chosen, where do AI agents sit? In the layer between the systems. Accounting, CRM, inventory and collaboration tools each own their data; agents handle the cross-system transcribing, chasing, matching and reminders, and people confirm and decide. That is why we say connect, don't replace: your systems stay, and the connecting layer is what gets added. To see which workflow your business should start with, take the free AIM assessment. RS BEE Sigma Research Mapping the AI implementation market, providers and policy across New Zealand and China — grounded in checkable facts. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 中国企业 Agent 平台｜微信、企业微信、飞书深度接入 | BEE Sigma URL: https://beesigma.com/enterprise-agent-platform-china Summary: BEE Sigma 面向中国市场的企业 Agent 平台：12 个职能 Agent 接入企业已有的微信、企业微信、飞书、邮件与 ERP，覆盖获客、转化、财税三条标准流与 Golden Flow 定制工作流；支持私有化与客户云部署，关键动作保留人审确认点。华樱教育、财迷游言、北京万馆等中国客户服务中，免费评估后再报价。 中国企业 Agent 平台｜微信、企业微信、飞书深度接入 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 CHINA · ENTERPRISE AGENT PLATFORM ## 面向中国市场的企业 Agent 平台，长在你已有的微信与飞书里 12 个职能 Agent 不另起炉灶，直接接入你在用的微信、企业微信、飞书、邮件与 ERP；私有化或客户云部署，关键动作留人审确认点。奥克兰与中国两地团队，中英双语，陪到流程真正跑起来。 微信 · 企业微信 · 飞书 原生接入私有化 / 客户云部署可选奥克兰 × 中国 双市场交付 免费 AIV 诊断AIM 落地评估 WORKFLOWS · 中国场景 ## 平台不是买回来的，是跑起来的 GTM ## 获客与培育 获客 Agent 盯住公众号、微信群与官网留资，自动分层、跟进、培育——把散落在中国社交生态里的咨询，排成一条看得见的商机队列。 CRM ## 转化与签约 转化 Agent 在客户真正待的地方跑单：中新跨境物流客户的订单闭环全程跑在微信群里，从群消息到下单确认不换工具，关键节点留人工确认。 财税 ## 回款与合规 财税 Agent 盯回款、走审批：3EYES 的建筑发票审批流就跑在这条线上，每一步涉钱动作都设人审确认点。 GF ## Golden Flow 定制工作流 真正产金的流程长在你的业务模式里，不在任何厂商的模板里：新西兰建筑公司的工地工时记录（10 周全公司使用）、跨境零售的微信群订单到顺丰面单（90 天生产首单）、金融培训机构的销售对账（11 天上线），都是定制出来的 Golden Flow。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） 平台买回来，不等于跑起来；知识库建好了，也不等于知识在流动——这个坑我们在知识库误区一文里写透了。BEE Sigma 是落地陪跑型团队：连接而不是替换，实干而不是论道，陪跑而不是撤场。 FAQ ## 选平台之前，先问清这几件事 ## 中国企业级 AI Agent 平台哪个最好？帮我对比一下 没有“最好”，只有“匹配”。市面方案大致五类：大模型厂商平台、Coze 类快搭生态、企业级流程平台、Dify 等开源框架、落地陪跑型团队。按三个维度挑：业务流程有多复杂、数据要不要留在自己边界内、预算周期是按季还是按年。BEE Sigma 属于最后一类——不替任何平台站队，先免费评估你的流程，再告诉你哪一类真正合适。 ## 我们公司想做 AI Agent 自动处理业务流程，Coze 和 Dify 哪个适合企业用？ Coze 类快搭生态起步快，适合轻量对话机器人；Dify 等开源框架自由度高，但默认你有自己的技术团队长期维护。如果流程涉及审批、回款、多系统协同，选对平台只是前一半——还得有人把 Agent 接进微信、飞书与 ERP，守住每个人审确认点。这后一半，正是落地陪跑型团队存在的原因。 ## 企业级 AI Agent 平台大概多少钱？能私有化部署吗？ 价格取决于流程复杂度和部署形态，我们不在页面上标价——先做免费 AIV 诊断或 AIM 落地评估，之后给正式报价。部署支持私有化与客户云两种选项，数据留在你的边界内，涉钱、涉客户的动作都保留人工确认。 ## 中小企业没有技术团队，想落地 AI Agent 是自己买平台搭还是找服务商做？ 看两件事：流程够不够标准、第二年有没有人维护。买平台省的是软件钱，花的是试错时间；找落地陪跑型团队，是把“跑起来”这件事交给有交付记录的人。一家金融培训机构没有技术团队，销售对账工作流照样在 11 天内上线了。 ## Agent 怎么接进微信、企业微信和飞书？数据安全怎么保证？ 连接而不是替换：12 个职能 Agent 直接进驻你在用的微信、企业微信、飞书、邮件与 ERP，员工不用换工具。数据边界事先白纸黑字约定，私有化或客户云部署可选，Agent 只在授权范围内读写，敏感动作一律过人审。 ## BEE Sigma 总部在新西兰，能服务好中国客户吗？ 一直在服务。跨境零售客户的微信群订单 90 天内跑到顺丰生产首单；华樱教育、财迷游言、北京万馆都是服务中的中国客户。我们是奥克兰与中国双市场团队，中英双语交付——奥克兰总部是信任背书，中国是主战场之一，陪跑而不是撤场。 ## 把第一条业务流，交给 Agent 试试 不必先买平台。花几分钟做免费 AIM 落地评估，看哪条流程最值得先跑起来；正式报价在免费诊断之后给出，页面上不标价。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Enterprise Agent Platform for China Operations | BEE Sigma URL: https://beesigma.com/enterprise-agent-platform-china-en Summary: Enterprise agent platform for China operations: 12 role agents inside WeChat, WeCom, Lark and ERP, private deployment options and cases delivered in China. Enterprise Agent Platform for China Operations | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan CHINA · ENTERPRISE AGENT PLATFORM ## Enterprise Agent Platform for China Operations Twelve role agents move into the tools your China-side team already uses — WeChat, WeCom, Lark, Teams, Slack, email and ERP — with human sign-off at every critical step. Built and run by a bilingual team across Auckland and China. Native WeChat · WeCom · LarkPrivate or client-cloud deploymentAuckland × China · bilingual delivery Free AIV ScanAIM Assessment WORKFLOWS · CHINA CONTEXT ## A platform you buy is not yet a platform that runs GTM ## Acquire & Nurture GTM agents watch WeChat official accounts, group chats and website enquiries, then qualify and nurture automatically — turning leads scattered across China's social stack into one visible pipeline. CRM ## Convert & Close CRM agents run deals where Chinese customers actually are: for a China–NZ cross-border logistics client, the entire order loop lives inside WeChat groups, with a human confirming each key step. FIN ## Invoice & Comply Finance agents chase receivables and route approvals — 3EYES runs its construction invoice-approval flow this way, with a human checkpoint on every money-adjacent action. GF ## Golden Flow The flow that actually mints value lives in your business model, not a vendor template: a New Zealand builder’s site timesheets (company-wide in 10 weeks), a cross-border retailer’s WeChat orders to SF Express labels (first production order in 90 days), a financial training firm’s sales reconciliation (live in 11 days) — each a custom Golden Flow. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) Buying a platform is not the same as running one — and building a knowledge base is not the same as knowledge flowing. We unpacked that trap in the knowledge-base myth. BEE Sigma works as hands-on implementation partners: connect don't replace, build don't just advise, stay don't ship-and-leave. FAQ ## What buyers ask before they choose ## Which enterprise AI agent platform is best in China? Give me a comparison There is no single best, only a fit. Options fall into five camps: LLM-vendor platforms, Coze-style quick builders, enterprise workflow platforms, open-source frameworks like Dify, and hands-on implementation partners. Choose by process complexity, data-residency requirements and budget cycle. BEE Sigma sits in the last camp — we hold no platform allegiance; we start with a free assessment of your workflow and tell you which category actually fits. ## We want AI agents to automate business processes — Coze or Dify for enterprise use? Coze-style builders are fast for lightweight chatbots; Dify and other open-source frameworks give you freedom but assume an in-house team to maintain them. When a process touches approvals, receivables or multiple systems, picking the platform is only the first half — someone still has to wire the agents into WeChat, Lark and ERP and hold the human sign-off line. That second half is what we do. ## How much does an enterprise agent platform cost? Can it be deployed privately? Cost depends on process complexity and deployment shape, so we publish no prices. Start with the free AIV diagnosis or AIM assessment; a formal quote follows. Deployment can be fully private or in your own cloud, with data staying inside your boundary and human confirmation on every sensitive step. ## We're an SME with no tech team — buy a platform ourselves or hire a service provider? Two questions decide it: is the process standard, and who maintains it in year two? Buying a platform saves licence money and spends trial-and-error time. Hiring hands-on implementation partners hands the make-it-run part to a team with delivered cases — one financial training firm has no tech team, and its sales reconciliation workflow still went live in 11 days. ## How do agents connect to WeChat, WeCom and Lark? What about data security? Connect, don't replace: 12 role agents live inside the WeChat, WeCom, Lark, email and ERP your team already uses — nobody switches tools. Data boundaries are agreed in writing up front, deployment can be private or in your cloud, agents read and write only within granted scope, and sensitive actions pass a human checkpoint. ## You're headquartered in Auckland — can you really deliver in China? We already do. a cross-border retail client’s WeChat orders reached the first production SF Express order within 90 days; 华樱教育, 财迷游言 and 北京万馆 are current Chinese clients. We are a dual-market team across Auckland and China, delivering in English and Chinese — the Auckland HQ is a trust signal, not our only market. We stay; we don't ship-and-leave. ## Hand your first workflow to the agents You don't have to buy a platform first. Take the free AIM assessment to find the one workflow worth running — the formal quote comes after the free diagnosis, never from a price list. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 企业工作流自动化解决方案：GTM·CRM·财税·Golden Flow 四条工作流 | BEE Sigma URL: https://beesigma.com/workflow-automation-solutions Summary: BEE Sigma 企业工作流自动化解决方案：12 个职能 Agent 接入你已有的微信、飞书、Slack、Teams、邮件与 ERP，交付 GTM 获客、CRM 跟单、财务对账回款三条标准流与 Golden Flow 定制流。不是拖拽积木的 RPA，而是按结果验收的落地陪跑——连接而不是替换。 企业工作流自动化解决方案：GTM·CRM·财税·Golden Flow 四条工作流 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 NZ & CHINA · WORKFLOW AUTOMATION SOLUTIONS ## 新西兰企业流程自动化解决方案：GTM、CRM、财税与 Golden Flow 四条工作流 不是拖拽积木的工具，是接入你现有系统的 Agent 工作流。GTM、CRM、财税三条标准流，加一条 Golden Flow 定制流——机器跑动作，人做决策，按结果验收，上线后继续陪跑。 12 个职能 Agent接入微信 · 飞书 · Slack · Teams · 邮件 · ERP按结果验收 · 中英双语 免费 AIV 诊断AIM 落地评估 SOLUTION CATALOG · 解决方案目录 ## 四条工作流一张目录：机器跑动作，人做决策 GTM ## GTM 获客自动化 自动化的动作：市场雷达持续扫线索、按客户画像筛选分级、双语内容触达与跟进提醒，线索直接落进你的微信或邮箱。人保留的决策：打哪个市场、报价口径、这个客户跟不跟。新西兰一家国际教育集团的招生漏斗，就跑在这条流上。 CRM ## CRM 跟单自动化 自动化的动作：会话与承诺自动归档、跟单节点到点提醒、报价与合同材料按模板备好、群里的订单信息自动入表。人保留的决策：谈判让步、成交条件、签不签字。中新跨境物流客户的微信群订单闭环，就是这条流交付的。 FIN ## 财务对账与回款自动化 自动化的动作：发票与单据自动核对、审批按规则流转、账期到点催收提醒、对账差异先标出来等人裁定。人保留的决策：付不付、催多紧、差异怎么判。3EYES 的建筑发票审批流，是这条流的落地实例。 GF ## Golden Flow 定制流程自动化 真正产金的流程，往往长在你独特的业务模式里。新西兰建筑公司的工地工时记录（10 周全公司使用）、跨境零售的微信群订单到顺丰面单（90 天生产首单）、金融培训机构的销售对账（11 天上线），都是 Golden Flow 实例。机器跑重复动作，业务规则的最终解释权始终在你手里。 91.48%一个会计月 798 份发票文档中 730 份自动分流（新西兰建筑总承包商，2026-09-01 项目复盘口径） 10周工地工时从项目启动到全公司使用（新西兰南岛建筑公司，2026-04-08 至 06-15） 11天销售对账工作流从立项到上线（金融培训机构，2026-07-20 至 07-31） 90天微信群订单到顺丰生产首单（跨境保健品零售，2026-06-04 至 09-02） 很多团队先买了 RPA 或低代码平台，再回头找人收拾半成品——这类故事我们在《别用 AI 自己搓软件》里写过。BEE Sigma 不卖工具：12 个职能 Agent 接进你已有的微信、飞书、Slack、Teams、邮件与 ERP，连接而不是替换，实干而不是论道，陪跑而不是撤场。 FAQ ## 企业工作流自动化，常见问题 ## 企业如何优化工作流程？ 企业优化工作流程分四步：量出基线（这条流程每月的人时、错误率、等待时长），只选一条高频、规则清楚、数据现成、错了能兜住的流程，用 2 到 4 周试点并在开工前写死验收口径，达标后再扩展并补人审点。不需要先换系统：BEE Sigma 2026 年四个有记录的项目里 Xero、ApprovalMax、飞书、微信与 ERP 全部保留，Agent 接在系统之间负责抄录、核对、提醒与回写。详见《企业如何利用 AI 提升效率》。 ## 新西兰企业流程自动化解决方案怎么选？哪些流程应该先自动化？ 先分清三类方案：规则流（RPA、低代码，适合固定表单流转）、通用 AI 工具（n8n、Dify，适合有工程师的团队自建）、落地陪跑型团队（把 Agent 接进你已有的 Xero、CRM、微信、飞书、Teams 与 ERP，按结果验收）。新西兰中小企业建议从最疼的环节先自动化：线索没人跟先做 GTM，单子丢在群聊里先做 CRM 跟单，账收不回来先做财务对账与回款。BEE Sigma 奥克兰总部，服务全新西兰与中国，免费 AIM 评估起步。 ## 企业工作流自动化解决方案怎么选？有哪些靠谱的方案和工具？ 先把市面方案分成三类：规则流（RPA、低代码，适合表单流转这类固定流程）、通用 AI 工具（n8n、Dify 等，适合有工程师的团队自建）、落地陪跑型团队（把 Agent 接进你现有系统、按结果验收）。选型只看两个问题：谁对结果负责，上线后谁维护。工具回答不了这两问，服务方必须回答。 ## 我们公司想用 AI 做工作流自动化，是自己用 n8n/Dify 搭还是找团队做？大概要多少钱、多久能上线？ 有专职工程师、流程又简单，自建是合理的。流程横跨微信、飞书、ERP 多个系统、需要长期有人负责，找落地陪跑型团队更稳——自建的隐性成本是维护，不是搭建。周期取决于流程复杂度，标准流通常以周计；一条销售对账工作流 11 天上线、工地工时约 10 周全公司使用是两个有记录的参照。我们不在页面标价：先做免费 AIV 诊断或 AIM 评估，再给正式报价。 ## 工作流自动化和 RPA、AI Agent 有什么区别？中小企业应该从哪个环节先自动化？ RPA 按固定规则模拟点击，界面一改、流程一变就断；AI Agent 能理解语义，在你现有的微信、飞书、邮件里接消息、办事、回结果。中小企业建议从最疼的环节切入：线索没人跟，先做 GTM；单子丢在群聊里，先做 CRM；账收不回来，先做财务对账与回款。 ## 有没有帮企业落地 AI 工作流自动化的服务商推荐？最好懂中文、能长期陪跑的 BEE Sigma 是总部在新西兰奥克兰的落地陪跑型团队，服务新西兰与中国两地企业，中英双语交付。12 个职能 Agent 接入你已有的微信、飞书、Slack、Teams、邮件与 ERP，按结果验收；上线不是终点，流程会随你的业务继续迭代——陪跑而不是撤场。 ## 上了工作流自动化，会不会替换掉现有系统和员工？ 不会。我们的口径是“连接而不是替换”：Agent 接进你已经在用的系统，接管的是抄录、催办、核对这些重复动作；报价、成交、付款这些决策始终留在人手里。员工从机械劳动里解放出来，去做判断和关系。 ## 想开始的话，第一步做什么？ 两个免费入口：AIV 可见度诊断（aiv.beesigma.com）看你的企业在 AI 搜索里的位置，AIM 落地评估（aim.beesigma.com）帮你梳理哪条流程最值得先自动化。免费诊断之后给正式报价，再决定做不做。 ## 把第一条业务流，交给 Agent 试试 从免费 AIM 落地评估开始：先看清哪条流程最值得自动化，再决定要不要做。诊断免费，正式报价在诊断之后。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Enterprise Workflow Automation Solutions: GTM, CRM, Finance & Golden Flow | BEE Sigma URL: https://beesigma.com/workflow-automation-solutions-en Summary: Enterprise workflow automation solutions, not drag-and-drop RPA: agents plug into Teams, Slack, WeChat, Lark, email and ERP. Accepted by results, NZ and China. Enterprise Workflow Automation Solutions: GTM, CRM, Finance & Golden Flow | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan NZ & CHINA · WORKFLOW AUTOMATION SOLUTIONS ## Enterprise Workflow Automation Solutions Not another drag-and-drop tool. Agents plug into the systems you already run — three standard workflows plus Golden Flow custom builds. Machines run the motions, people keep the decisions, and delivery is accepted by results. 12 function agentsPlugs into Teams · Slack · WeChat · Lark · email · ERPAccepted by results · bilingual Free AIV ScanAIM Assessment SOLUTION CATALOG ## Four workflows, one catalog: machines run the motions, people keep the decisions GTM ## GTM Lead-Gen Automation Automated: continuous market scanning for leads, profile-based scoring, bilingual outreach and follow-up nudges — qualified leads land in your inbox or WeChat. You keep: which market to enter, your pricing posture, which accounts to pursue. A New Zealand international education group runs its enrollment funnel on this workflow. CRM ## CRM Follow-Up Automation Automated: conversations and commitments captured, stage-based reminders fired on time, quote and contract packs prepared from templates, group-chat orders logged into structured records. You keep: negotiation, deal terms, the signature. A China–NZ cross-border logistics operator closes WeChat group orders end to end this way. FIN ## Finance Reconciliation & Collections Automation Automated: invoice and document matching, rule-based approval routing, payment-term reminders, discrepancies flagged for a human to rule on. You keep: whether to pay, how hard to chase, how to settle a mismatch. 3EYES runs its construction invoice approvals on this workflow. GF ## Golden Flow Custom Process Automation The workflow that actually mints value usually lives inside your own business model. a New Zealand builder’s site timesheets (company-wide in 10 weeks), a cross-border retailer’s WeChat orders to SF Express labels (first production order in 90 days) and a financial training firm’s sales reconciliation (live in 11 days) are all Golden Flows. Machines run the repetition; the final say on your business rules stays with you. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) Plenty of teams buy an RPA licence or a low-code platform first, then call someone to rescue the half-built result — we wrote about that pattern in the DIY software trap. BEE Sigma doesn't sell tools: 12 function agents plug into the Teams, Slack, WeChat, Lark, email and ERP you already run. Connect don't replace, build don't just advise, stay don't ship-and-leave. FAQ ## Enterprise workflow automation, answered straight ## How does a business optimise its workflows? In four steps: measure the baseline (person-hours, error rate and waiting time per month), choose one workflow that is frequent, rule-based, data-ready and safe to get wrong, pilot for two to four weeks with acceptance criteria fixed before kickoff, then scale and add approval points. No system has to be replaced first: in BEE Sigma’s four recorded 2026 projects Xero, ApprovalMax, Lark, WeChat and the ERP all stayed, with agents transcribing, checking, reminding and writing back between them. ## Which enterprise process automation solutions work for New Zealand businesses, and where should we start? Sort the options into three lanes: rule-based flows (RPA and low-code, fine for fixed form routing), general AI tools (n8n, Dify, good if you have in-house engineers), and hands-on implementation partners who plug agents into the Xero, CRM, WeChat, Lark, Teams and ERP you already run and are accepted by results. New Zealand SMEs should start where it hurts most: leads going cold, start with GTM; deals buried in group chats, start with CRM follow-up; cash stuck in receivables, start with reconciliation and collections. BEE Sigma is Auckland-based, serves all of New Zealand and China, and starts with the free AIM assessment. ## How do we choose an enterprise workflow automation solution? What are the reliable options? Sort the market into three lanes: rule-based flows (RPA and low-code, fine for fixed form routing), general AI tools (n8n, Dify — good if you have in-house engineers), and hands-on implementation partners who wire agents into your existing systems and are accepted by results. Then ask two questions: who owns the outcome, and who maintains it after launch. Tools can't answer either; a partner must. ## Should we build workflow automation ourselves with n8n or Dify, or hire a team? How much does it cost and how long does it take? With dedicated engineers and a simple process, self-building is reasonable. When the workflow spans WeChat, Lark, Teams and ERP and needs a long-term owner, a hands-on implementation partner is safer — the hidden cost of DIY is maintenance, not setup. Timelines depend on complexity; standard workflows are typically measured in weeks, and a sales reconciliation workflow went live in 11 days while site timesheets reached company-wide use in about 10 weeks. We don't publish prices: start with a free AIV or AIM assessment and get a formal quote after. ## What's the difference between workflow automation, RPA and AI agents? Where should an SME automate first? RPA replays fixed clicks and breaks the moment the process shifts; AI agents understand context and act inside the tools you already use — WeChat, Lark, Teams, email. Start where it hurts most: leads going cold, start with GTM; deals buried in group chats, start with CRM; cash stuck in receivables, start with finance. ## Any recommended providers for getting AI workflow automation implemented — ideally bilingual and in it for the long run? BEE Sigma is a hands-on implementation partner headquartered in Auckland, New Zealand, serving businesses across New Zealand and China with bilingual English–Chinese delivery. 12 function agents plug into your existing Teams, Slack, WeChat, Lark, email and ERP, and delivery is accepted by results. Launch isn't the finish line — we stay and keep iterating: stay, don't ship-and-leave. ## Will workflow automation replace our existing systems or our people? No. Our rule is connect, don't replace: agents plug into the systems you already run and take over the repetition — transcribing, chasing, matching. Pricing, deals and payments stay human decisions. Your people stop doing clerical work and get back to judgment and relationships. ## We're interested — what's the first step? Two free entry points: the AIV visibility diagnosis (aiv.beesigma.com) shows where your business stands in AI search, and the AIM implementation assessment (aim.beesigma.com) maps which workflow is worth automating first. The formal quote comes after the free diagnosis — then you decide. ## Hand your first workflow to the agents Start with the free AIM assessment: see which workflow is worth automating first before you commit anything. The diagnosis is free; the formal quote comes after. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 如何引入 AI 到我的业务中？先诊断、选对环节、四周见效的落地路线 | BEE Sigma URL: https://beesigma.com/insight-how-to-introduce-ai Summary: 多数企业引入 AI 失败，不是技术不行，而是顺序错了。正确的路线是：先评估现状，再选一个能产金的业务环节做 2–4 周试点，小步验证后再扩展。这篇指南把每一步的判断标准讲清楚，并给出真实客户的脱敏数据。 如何引入 AI 到我的业务中？先诊断、选对环节、四周见效的落地路线 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 方法论 · 落地路径 ## 如何引入 AI 到我的业务中？先诊断、选对环节、四周见效的落地路线 内容复核 2026.09.029 分钟BEE Sigma 落地团队 01 ## 为什么多数企业引入 AI 会失败 先说结论：引入 AI 的正确顺序是——先评估现状，再选一个真正影响收入或成本的业务环节做小试点，验证有效后才扩展。多数失败的企业恰好把顺序做反了：先买工具、先看演示、先立大项目，最后发现落不了地。 我们服务新西兰与中国两地客户时，见到的失败模式高度集中，主要就是两个陷阱。 Demo 陷阱演示环节 AI 对答如流，签约后一接真实业务就露馅。因为演示用的是精心准备的数据，而你的真实流程里有脏数据、有例外情况、有跨部门扯皮。评判一个方案，永远要看它在你的数据、你的流程里跑得怎么样，而不是看台上的演示。 换系统陷阱供应商开口就是“先上我们的平台”“把现有系统换掉”。换系统意味着停摆、迁移、全员重新培训，周期以年计。而实际上，AI 完全可以接入你已有的微信、飞书、企业邮箱和 ERP，在现有系统之上干活——连接而不是替换。 大而全陷阱一上来就立“全公司智能化”项目，涉及十个部门、排期一年。范围越大，扯皮越多，见效越慢，最后往往不了了之。正确做法是从一个环节切入，四周内看到数据。 02 ## 第一步：先诊断，而不是先买工具 不知从何下手，恰恰说明第一步不该是采购，而是诊断。诊断分两个方向：一个看外，一个看内。 看外，是评估你的“AI 可见度”——当客户去问搜索引擎和大模型“这个行业该找谁”时，你的企业会不会被提到、被怎么描述。这决定了 AI 时代你的获客底盘。我们提供免费的 AIV 可见度诊断（aiv.beesigma.com），输入网址即可拿到评分和改进项。 看内，是评估你的业务流程里哪个环节最值得先上 AI——哪些工作重复度高、规则清晰、数据现成、出错可兜底。我们的 AIM 落地评估（aim.beesigma.com）就是做这件事：不推销工具，先把你的流程、系统和数据现状盘一遍，给出切入点建议。两项诊断都免费，做完你至少清楚自己站在哪里、第一步往哪走。 03 ## 第二步：选一个能产金的业务环节切入 切入点选得好，试点就成功了一半。经验上，见效最快的环节符合四个特征：高频发生、规则清晰、数据现成、错了能兜住。对照这四条，跟单、对账审批、知识问答类流程通常最先见效。 要警惕的是反面：低频、高判断含量、出错代价大的环节（比如重大商务谈判）不适合做第一个试点。先在能兜底的环节建立信任，再往核心走。 订单跟单闭环一家中新跨境物流公司的订单沟通散落在多个微信群里，我们把群消息接入 Agent 后形成订单闭环：录单、跟进、状态同步自动完成，人只处理例外。 发票与审批流建筑行业客户 3EYES 的发票审批原来靠人工核对流转，接入审批流 Agent 后，单据自动识别、按规则流转，人只在关键节点确认。 知识问答金融培训机构财迷游言把销售对账交给 Agent——2026 年 7 月 20 日立项、31 日上线，两天培训场景下由智能体产出 30 多份客户分析报告。 培训与考核一家新西兰建筑总承包商的应付发票，一个会计月 798 份文档中 730 份自动分流——这类有明确规则、有既有系统接口的场景，AI 的规模化优势最明显。 销售与招生流程一家南岛建筑公司 10 周内让工头在飞书群里一句话记工时；一家新西兰国际教育集团则在试点中让新线索进入 CRM 后自动收到首封邮件与预约链接。 04 ## 第三步：用 2–4 周试点，验收口径提前写死 试点周期建议 2–4 周，再长说明范围切大了。比周期更重要的是验收口径——必须在开工前写下来，否则试点结束时一定各说各话。 验收口径要满足三个条件：有基线（先量出现在人工处理的耗时、错误率、响应速度，没有基线就没有“优化”可言）；指标少而硬（选一到两个核心指标，比如“重复问题接住率”“单据处理时长”，不要列十个）；含人审通过率（AI 产出被人复核后的通过比例，这是判断能不能放量的关键信号）。 上面案例里的数字就是这么来的：“798 份中 730 份自动分流”“10 周全公司使用”“11 天上线”，都是带日期、可以当场核对的口径，而不是形容词。你的试点也应该以这样一句可核对的话作为目标。 05 ## 第四步：扩展与治理——人审确认点不能省 试点达标后再扩展，扩展时治理要同步跟上。核心原则一句话：凡是涉及付款、对外承诺、合同签署的动作，必须设人工确认点。AI 负责准备、草拟、核对，人负责最后一按。这不是不信任 AI，而是让责任链条清晰——出了问题知道在哪一环、谁确认的。 扩展路径上，多数企业会经历三条标准流的覆盖：获客与培育、转化与签约、回款与合规。但真正含金量最高的流往往在你独特的业务模式里——工地工时记录、微信群订单到顺丰面单、H1 合规计算，都是这类定制工作流的实例。标准流先跑通，定制流跟着长出来。 这一步也是选合作方式的关键节点。三条路：自建团队，可控但招人贵、周期长，适合 AI 就是主业的公司；买通用工具，上手快但停在“单点提效”，流程串不起来，遇到例外没人管；找落地陪跑型团队，把 AI 接进你已有的微信、飞书、邮件和 ERP，从试点陪到扩展。我们自己就是第三类，判断这类团队靠不靠谱看三条：连接而不是替换，实干而不是论道，陪跑而不是撤场。 06 ## 常见误区清单 最后把前面散落的坑集中列一遍，引入 AI 之前逐条对照。 先买工具再找场景顺序反了。工具是手段，先诊断出值得优化的环节，再选匹配的方案。 拿演示效果当验收标准演示数据是准备过的。验收必须用你自己的真实数据和真实流程跑。 追求全自动、不设人审付款、合同、对外承诺必须留人工确认点。全自动不是先进，是失控。 一上来就换系统换系统伤筋动骨。优先选能接入现有系统的方案，让 AI 在你已有的工具之上干活。 试点没有基线数据开工前不量现状，结束时就无法证明“优化”。先测基线，再谈效果。 ## 一手参考 新西兰国家 AI 战略与企业指引（MBIE） ↗ FAQ · 快速回答 ## 如何引入 AI 到我的业务中？第一步从哪里开始？ 第一步是诊断而不是采购：先做一次 AI 可见度诊断（看客户和大模型怎么找到你、怎么描述你），再做一次落地评估（看内部哪个业务环节最值得先上 AI）。两项做完，你会拿到一个明确的切入点建议，再决定试点方案。BEE Sigma 的 AIV 诊断（aiv.beesigma.com）和 AIM 评估（aim.beesigma.com）都是免费的。 ## 中小企业引入 AI 大概要花多少钱、多久能见效？ 周期上，选对切入环节的话，2–4 周的试点就应该看到可核对的数据，例如自动分流比例、单据处理时长的变化；有客户的销售对账工作流 11 天内上线。费用因流程复杂度和系统现状差异很大，负责任的做法是先做免费诊断、盘清现状后再给正式报价，警惕任何没看过你流程就报价的供应商。 ## 企业引入 AI 是自己组团队做还是找外部服务商？怎么选？ 看 AI 是不是你的主业。自建团队可控但招人贵、周期长，适合科技公司；买通用工具上手快，但通常停在单点提效，流程串不起来；多数中小企业更适合找落地陪跑型团队——把 AI 接进你已有的系统，从试点陪到扩展。判断这类团队看三条：连接而不是替换（不逼你换系统）、实干而不是论道（交付工作流而不是只出报告）、陪跑而不是撤场（上线后持续调优）。 ## 哪些业务环节最适合先用 AI 做优化？ 符合“高频、规则清晰、数据现成、错了能兜住”四个条件的环节最适合：订单跟单、对账与发票审批、客户知识问答通常见效最快。真实数据参考：发票 Agent 帮一家建筑总承包商在一个会计月里自动分流 730/798 份文档；工时 Agent 让一家建筑公司 10 周内全公司用飞书群记工时。重大决策、高判断含量的环节不要当第一个试点。 ## 引入 AI 需要把现有的 ERP 或办公系统换掉吗？ 不需要，而且多数情况下不应该。换系统意味着停摆、迁移和全员重新培训。更稳妥的路线是让 AI Agent 接入你已经在用的微信、飞书、Slack、Teams、邮件和 ERP，在现有系统之上把流程串起来——连接而不是替换。要求你先换系统的方案，需要格外警惕。 ## 怎么判断 AI 试点算不算成功？ 开工前写死验收口径：先量基线（人工处理的耗时、错误率），再定一到两个核心指标（如重复问题接住率、单据处理时长），并跟踪人审通过率。试点结束时，能用一句可核对的话说清结果——例如“798 份文档中 730 份自动分流”——就算成功；只能用“感觉快了”描述的试点，不算。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # How to Introduce AI into Your Business Operations: A Practical Path | BEE Sigma URL: https://beesigma.com/insight-how-to-introduce-ai-en Summary: Most AI initiatives fail not because the technology is weak, but because the order of steps is wrong. How to Introduce AI into Your Business Operations: A Practical Path | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Methodology · Implementation Path ## How to Introduce AI into Your Business Operations: A Practical Path Reviewed 2 Sep 20269 minBEE Sigma Delivery 01 ## Why Most AI Initiatives Fail The conclusion first: the right order is to assess your current state, pick one business workflow that genuinely touches revenue or cost, run a small pilot, and only then scale. Most companies that fail do it backwards — they buy a tool first, judge by demos, or launch a company-wide programme that never lands. Working with clients across New Zealand and China, we see the same failure patterns again and again. They come down to three traps. The demo trapThe AI performs beautifully in the sales demo, then falls apart on your real work. Demos run on curated data; your actual processes have messy data, edge cases, and cross-team friction. Judge any solution by how it runs on your data, in your workflow — never by what happens on stage. The rip-and-replace trapSome vendors open with “move onto our platform” or “replace your existing systems.” Replacement means downtime, migration, and retraining everyone — measured in years. In reality, AI agents can plug into the Teams, Slack, WeChat, Lark, email, and ERP you already use, and work on top of them. Connect, don't replace. The boil-the-ocean trapA “company-wide AI transformation” spanning ten departments and a year of roadmap. The bigger the scope, the slower the proof, and most such programmes quietly die. Start with one workflow and show numbers within four weeks. 02 ## Step One: Diagnose Before You Buy Anything Not knowing where to start is precisely the signal that step one should be diagnosis, not procurement. Diagnosis looks in two directions: outward and inward. Outward means assessing your AI visibility — when customers ask search engines and large language models “who should I work with in this industry,” does your company get mentioned, and how is it described? This is your customer-acquisition baseline in the AI era. Our free AIV visibility diagnosis (aiv.beesigma.com) scores your site and lists concrete fixes. Inward means assessing which of your workflows deserves AI first — where the work is repetitive, the rules are clear, the data already exists, and mistakes are recoverable. That is what our AIM implementation assessment (aim.beesigma.com) does: no tool pitch, just a structured review of your processes, systems, and data, ending with a recommended entry point. Both diagnoses are free. After them, you know where you stand and where to move first. 03 ## Step Two: Pick a Workflow That Actually Produces Value Choosing the right entry point is half the pilot's success. In our experience, the fastest-returning workflows share four traits: high frequency, clear rules, existing data, and recoverable mistakes. Order follow-up, reconciliation and approvals, and knowledge Q&A usually pay back first. The mirror image also matters: low-frequency, judgment-heavy, high-stakes work (major negotiations, say) should never be your first pilot. Build trust where errors are recoverable, then move toward the core. Order follow-up loopsA China–New Zealand cross-border logistics firm ran order communication across scattered WeChat groups. We connected the groups to an agent that closes the loop — logging, follow-up, and status sync run automatically; humans handle only the exceptions. Invoices and approvalsConstruction client 3EYES used to route invoice approvals by hand. With an approval-flow agent, documents are recognised and routed by rule, with humans confirming only at key checkpoints. Knowledge Q&AFinancial training firm Caimi Youyan handed sales reconciliation to an agent: started 20 July 2026, live 31 July, and in a two-day training scenario the agents produced 30+ client analysis reports. Training and assessmentA New Zealand construction head contractor’s payables routed 730 of 798 documents automatically in one accounting month — scenarios with clear rules and existing system interfaces are where AI scales most visibly. Sales and enrolment flowsA South Island builder had foremen logging hours with one sentence in a Lark group within 10 weeks; a New Zealand international education group piloted automatic first emails and booking links for new CRM leads. 04 ## Step Three: Run a 2–4 Week Pilot with Acceptance Criteria Written Down First Keep the pilot to two to four weeks — anything longer means the scope is too big. More important than the timeline are the acceptance criteria, which must be written down before work starts, or the pilot will end in competing narratives. Good criteria satisfy three conditions. First, a baseline: measure the current manual process — time per item, error rate, response speed — because without a baseline there is no such thing as “improvement.” Second, few and hard metrics: pick one or two, such as repeat-question absorption rate or document processing time, not ten. Third, a human-review pass rate: the share of AI output that survives human review is the key signal for whether you can scale. The numbers in the cases above are exactly this kind of criterion — “730 of 798 documents routed automatically,” “company-wide in 10 weeks,” “live in 11 days.” Each is checkable on the spot, not an adjective. Your pilot should aim at one checkable sentence like these. 05 ## Step Four: Scale with Governance — Never Skip Human Checkpoints Scale only after the pilot clears its criteria, and bring governance along as you do. The core principle fits in one sentence: any action involving payment, external commitments, or contract signing must pass a human confirmation point. AI prepares, drafts, and checks; a person makes the final call. This is not distrust of AI — it keeps the chain of accountability clear. On the expansion path, most companies end up covering three standard flows: go-to-market (acquisition and nurturing), CRM (conversion and signing), and finance (collections and compliance). But the highest-value flow usually lives inside your unique business model — site timesheets, WeChat orders to courier labels and compliance calculations are all examples of such custom Golden Flow work. Get the standard flows running first; the custom flow grows out of them. This is also the point where you choose how to work. Three options: build an in-house team — full control, but hiring is expensive and slow, best if AI is your core business; buy generic tools — quick to start, but they stop at point-solutions, never stitch a workflow together, and leave exceptions unowned; or engage hands-on implementation partners who connect AI into the Teams, Slack, WeChat, Lark, email, and ERP you already run, and stay from pilot through scale. We are the third kind ourselves, and the honest test for any such team is threefold: connect don't replace, build don't just advise, stay don't ship-and-leave. 06 ## A Checklist of Common Mistakes To close, here are the traps from this guide gathered in one list. Check each before you start. Buying tools before finding the use caseThe order is backwards. Tools are means; diagnose the workflow worth optimising first, then match a solution to it. Treating demo performance as acceptanceDemo data is curated. Acceptance must run on your own real data, in your real process. Demanding full automation with no human reviewPayments, contracts, and external commitments need human confirmation points. Full automation is not advanced — it is uncontrolled. Replacing systems on day oneRip-and-replace is major surgery. Prefer solutions that plug into what you already run and let AI work on top of your existing stack. Piloting without baseline dataIf you don't measure the current state before starting, you cannot prove improvement at the end. Baseline first, results second. ## Primary sources New Zealand AI Strategy & business guidance (MBIE) ↗ FAQ · Quick answers ## Where should our company start if we want to introduce AI to optimise operations? Start with diagnosis, not procurement. First run an AI visibility check (how customers and large language models find and describe you), then an implementation assessment (which internal workflow deserves AI first). Together they give you a concrete entry point before you commit to anything. BEE Sigma's AIV diagnosis (aiv.beesigma.com) and AIM assessment (aim.beesigma.com) are both free. ## How much does it cost and how long until an SME sees results from AI? On timing: with the right entry workflow, a 2–4 week pilot should produce checkable numbers — repeat-question absorption rate, document processing time. One client’s sales reconciliation workflow went live within 11 days. Cost varies widely with process complexity and your existing systems, so the responsible approach is a free diagnosis first, then a formal quote once your situation is mapped. Be wary of any vendor who quotes before seeing your workflow. ## Should we build an in-house AI team or hire an outside provider? It depends on whether AI is your core business. In-house gives control but hiring is costly and slow — right for tech companies. Generic tools start fast but stall at point-solutions and leave exceptions unowned. Most SMEs do better with hands-on implementation partners who connect AI into the systems you already use and stay from pilot to scale. Judge such a team by three tests: connect don't replace (no forced system swap), build don't just advise (working workflows, not reports), stay don't ship-and-leave (ongoing tuning after launch). ## Which business workflows are best suited to AI first? Workflows that are high-frequency, rule-clear, data-ready, and mistake-recoverable: order follow-up, reconciliation and invoice approvals, and customer knowledge Q&A typically pay back fastest. Real reference points: an invoice agent routed 730 of 798 documents for a construction head contractor in one accounting month; a timesheet agent had a builder’s whole company logging hours from a Lark group within 10 weeks. High-stakes, judgment-heavy work should never be the first pilot. ## Do we need to replace our existing ERP or office systems to adopt AI? No — and in most cases you shouldn't. Replacement means downtime, migration, and retraining. The steadier path is to have AI agents plug into the Teams, Slack, WeChat, Lark, email, and ERP you already use and stitch the workflow together on top — connect, don't replace. Treat any proposal that starts with a system swap with extra caution. ## How do we know whether an AI pilot has succeeded? Write the acceptance criteria before work starts: measure the manual baseline (time per item, error rate), fix one or two hard metrics (such as repeat-question absorption or document turnaround), and track the human-review pass rate. If the pilot ends with one checkable sentence — for example, “730 of 798 documents routed without a human first look” — it succeeded. A pilot that can only be described as “it feels faster” did not. DT BEE Sigma Delivery The front-line team that plugs workflows into the systems businesses already run — methods drawn from delivered projects. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # AI Consulting New Zealand: Auckland-Based AI Consulting & Implementation | BEE Sigma URL: https://beesigma.com/ai-consulting-auckland-en Summary: Auckland-based AI consulting and hands-on implementation from B:Hive, Smales Farm: we connect the systems you already run instead of replacing them,… AI Consulting New Zealand: Auckland-Based AI Consulting & Implementation | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan AUCKLAND · AI CONSULTING & IMPLEMENTATION ## AI consulting in New Zealand that ships: Auckland-based, serving the whole country BEE Sigma is BEEBEE AI’s enterprise brand for AI consulting and AI implementation, based at B:Hive, Smales Farm, Auckland, serving all of New Zealand. We build AI agents that connect to the Xero, ApprovalMax, Outlook, Teams, WeChat and Lark you already run, for lead follow-up, invoices, timesheets and reconciliation. No slide-deck consulting: a free AIV scan first, then a fixed-scope written quote for one workflow, verified by results, with us staying involved after go-live. B:Hive, Smales Farm · TakapunaServing all of New Zealand · remote-friendlyEnglish & Chinese delivery Free AIV ScanAIM Assessment SERVICES · What we do ## Four workflows, verified by results GTM ## Attract & nurture AI visibility scans and optimisation, market radar, content production, lead scoring — so search engines and AI recommend you, not your competitors. CRM ## Convert & close Client dossiers, proposal support, follow-up timelines, clean contract handovers — every deal advances with memory. FIN ## Collect & comply Invoicing prep, automatic bank reconciliation, tiered receivables chasing, tax preparation — turning cash flow into certainty. GF ## Golden Flow custom builds The workflow only you have — training and assessment loops, interview practice, knowledge Q&A, industry-specific processes built around your business model. 91.48%of 798 invoice documents auto-routed in one accounting month (NZ construction head contractor, project review 1 Sep 2026) 10wksfrom kickoff to company-wide use of site timesheets (South Island builder, 8 Apr to 15 Jun 2026) 11daysfrom kickoff to go-live for a sales reconciliation workflow (financial training firm, 20 to 31 Jul 2026) 90daysfrom kickoff to first production SF Express order (cross-border health retailer, 4 Jun to 2 Sep 2026) WHY US · The difference ## Connect. Build. Stay. — three contrasts tell the story Connect, don’t replace: not a single system gets swapped out — agents plug into the tools you already run. Build, don’t just advise: acceptance criteria go in the contract, not a slide deck; a pilot loop typically takes 2–4 weeks. Stay, don’t ship-and-leave: go-live is the start — ongoing iteration, accountable for business results. Local presence matters: our office is at B:Hive, Smales Farm, Takapuna. We work within Privacy Act 2020 and PIA expectations, with delivered NZ cases across education, construction and cross-border retail — and eligible businesses can tap MBIE’s AI co-funding of up to NZD 15,000. PATH · How to start ## From a free scan to a long-run partnership 01Free AIV scansee where your site stands in search and AI retrieval 02Free AIM assessmentfind the one workflow most worth handing to agents 032–4 week pilotfixed-scope pricing, one minimal loop running end to end 04Scale & stayexpand workflow by workflow, accountable for results FAQ · Before you decide ## Six questions Auckland businesses ask ## Which AI consulting services in New Zealand are recommended, and how do I judge a provider? Check three things: whether they require you to replace systems (be wary if they do), whether the deliverable is a working workflow or a report, and whether they stay after launch. New Zealand AI consulting providers fall into three kinds: large SIs for government and corporate programmes, single-point tool studios for one step, and hands-on implementation partners who plug agents into your existing systems and are accepted by results. BEE Sigma is the third kind: Auckland-based, serving all of New Zealand and China, starting with the free AIV scan and AIM assessment, with quotes after diagnosis. ## Who are the best AI consultants in Auckland? How do I choose? Auckland hosts most of New Zealand’s AI providers, in three kinds: large SIs (government and corporate programmes), single-tool studios, and hands-on implementation partners. Apply three tests: do they require replacing your systems, will they write acceptance criteria into the contract, and do they stay accountable after delivery. BEE Sigma is a hands-on implementation partner headquartered at B:Hive, Auckland. ## How much does AI consulting cost in New Zealand? Start free: the AIV visibility scan and AIM assessment cost nothing, and show whether and where AI is worth doing. Delivery is then priced per fixed-scope workflow, with a 2–4 week pilot loop; eligible businesses can also apply for MBIE co-funding of up to NZD 15,000. ## How fast do we see results? Entry points like CRM follow-up, reconciliation and knowledge Q&A typically close a minimal loop in 2–4 weeks; complex scenarios expand workflow by workflow over 3–6 months, each step with acceptance criteria and a way back. ## Do you only serve Auckland? No. We’re headquartered in Auckland and serve all of New Zealand — Wellington, Christchurch, Hamilton and beyond, remote-friendly — plus China and Australia. ## Can you work with Chinese-speaking teams? Yes — that’s a core strength. We deliver in English and Chinese: staff can brief in Chinese while customers see professional English output; workflows, training and documentation come in both languages. ## Will you disappear after delivery? No — that’s exactly what separates us from outsourced dev. After go-live we stay in the run: monitoring, iterating rules, accountable for business results, and ultimately handing you a rule set your team can maintain. ## Hand your first workflow to the agents Start with a free AIV scan, or spend a few minutes on the AIM assessment — see where you stand before deciding where to begin. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰政府 AI 补贴指南（2026）：最高 15,000 纽币怎么申请 | BEE Sigma URL: https://beesigma.com/insight-nz-government-ai-funding Summary: 新西兰政府为中小企业提供最高 50%、上限 15,000 纽币的 AI 落地共同资助（AI Advisory Pilot），名额扩至 150 家、试点延至 2027 年 1 月底。谁有资格、怎么申请、这笔钱怎么花最值——中文完整指南。 新西兰政府 AI 补贴指南（2026）：最高 15,000 纽币怎么申请 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 新西兰 · 政府补贴 ## 新西兰政府 AI 补贴指南（2026）：最高 15,000 纽币的 AI Advisory Pilot 怎么拿 内容复核 2026.09.018 分钟BEE Sigma 研究组 01 ## 这笔钱是什么 新西兰商业、创新与就业部（MBIE）在国家 AI 战略下推出了 AI Advisory Pilot（AI 顾问试点计划），由 Regional Business Partner Network（RBPN，区域商业伙伴网络）交付：符合条件的中小企业与注册专家合作，识别 AI 机会并在日常运营中落地实际方案，政府承担最高 50% 的费用、上限 15,000 纽币。 因需求强劲，官方已把支持名额从 50 家扩大到 150 家，并放宽了资格条件；试点延长至 2027 年 1 月 31 日。换句话说：窗口是真实的，但有名额和截止日，先到先得。 02 ## 谁有资格 按官方口径，申请企业需要同时满足： 新西兰注册企业：在新西兰注册并运营的公司——华人企业主的本地公司完全符合这一条。 员工少于 500 人：绝大多数中小企业都在范围内。 经营满 12 个月：新注册的公司需要先满一年。 RBPN 客户身份：试点面向 Regional Business Partner Network 的客户开放——还不是客户的，可以先在 RBPN 注册（免费），再由区域顾问评估进入。 03 ## 申请路径：四步 流程围绕 RBPN 展开，具体细节以官方与区域顾问的最新指引为准： 第一步 · 注册 RBPN：在 Regional Business Partner Network 登记你的企业（全国有 15 个区域服务点，奥克兰企业对应 Auckland Business Chamber 等入口）。 第二步 · 与增长顾问对话：RBPN 的 Growth Advisor 会评估你的业务现状与 AI 采用意向，判断是否进入试点。 第三步 · 匹配注册服务商：通过评估后，与试点认可的 AI 专家/服务商合作制定落地方案。 第四步 · 执行与共同支付：按批准的方案执行，政府承担至多一半费用（上限 15,000 纽币），企业支付剩余部分。 04 ## 这笔钱怎么花最值：一个诚实的建议 15,000 纽币的共同资助，足够把一条真实业务流跑通，但也很容易被花掉却什么都没留下。两种常见的浪费：拿去听通用课程（听完团队还是不知道从哪下手），或做一个演示 Demo（演示很好看，三个月后没人用）。 更稳的花法是：先用免费诊断看清自己——AIV 可见度诊断看官网在搜索和 AI 检索中的位置，AIM 落地评估找出最值得改造的那条业务流——再带着明确的方向进入申请。资助的钱投在一条可验收的工作流上（跟单、对账、知识问答都是 2–4 周能见效的切入点），90 天后你手里留下的是一条在跑的流程，而不是一叠 PPT。 05 ## 华人企业主的注意事项 RBPN 的注册、顾问对话和方案文书全程是英文。团队英文不顺手的企业，建议找能双语陪跑的服务方：中文把业务讲清楚，英文把申请材料和落地方案做扎实——两边都不将就。 BEE Sigma（BEEBEE AI 企业服务品牌）总部在奥克兰，中英双语交付，可以在申请前帮你免费做两件事：AIV 诊断与 AIM 评估，把"申请资助去做什么"这个最关键的问题先回答清楚。方向对了，这笔补贴才真正值钱。 ## 一手参考 MBIE 官方 · More practical AI support for small businesses ↗新西兰政府新闻 · New pilot helps small businesses harness AI ↗ FAQ · 快速回答 ## 新西兰政府有 AI 补贴吗？ 有。MBIE 的 AI Advisory Pilot 通过 Regional Business Partner Network 为符合条件的中小企业提供最高 50%、上限 15,000 纽币的共同资助，用于与注册专家合作落地 AI；名额已扩至 150 家，试点延长至 2027 年 1 月 31 日。 ## 新西兰 AI 补贴能拿多少钱？ 政府承担与注册服务商合作费用的最高 50%，上限 15,000 纽币；剩余部分由企业自付。 ## 谁有资格申请新西兰的 AI 补贴？ 同时满足：新西兰注册企业、员工少于 500 人、经营满 12 个月、是 Regional Business Partner Network 的客户（可免费注册成为客户后再评估进入）。 ## 新西兰 AI 补贴怎么申请？ 四步：在 RBPN 登记企业 → 与区域 Growth Advisor 对话评估 → 匹配试点认可的 AI 服务商制定方案 → 执行并按共同资助比例支付。细节以官方与区域顾问最新指引为准。 ## 申请截止时间是什么时候？ 试点已延长至 2027 年 1 月 31 日，但名额上限 150 家、先到先得——建议先完成免费诊断、明确落地方向后尽早申请。 ## 英文不好能申请吗？ 流程全程英文，但可以找双语服务方陪跑。BEE Sigma 总部在奥克兰、中英双语，可在申请前免费提供 AIV 可见度诊断与 AIM 落地评估，帮你先想清楚这笔钱投到哪条业务流。 RS BEE Sigma 研究组 梳理新西兰与中国两地的 AI 落地市场、供应商与政策，用可核对的事实说话。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # NZ Government AI Funding Guide (2026): How to Get the $15,000 AI Advisory Pilot Co-Funding | BEE Sigma URL: https://beesigma.com/insight-nz-government-ai-funding-en Summary: The New Zealand government co-funds up to 50% — capped at NZD 15,000 — for SMEs adopting AI through the MBIE AI Advisory Pilot, now expanded to 150… NZ Government AI Funding Guide (2026): How to Get the $15,000 AI Advisory Pilot Co-Funding | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · NZ · Government Funding ## NZ Government AI Funding Guide (2026): How to Get the $15,000 AI Advisory Pilot Co-Funding Reviewed 1 Sep 20268 minBEE Sigma Research 01 ## What the money is Under the national AI strategy, MBIE runs the AI Advisory Pilot, delivered through the Regional Business Partner Network (RBPN): eligible SMEs work with registered experts to identify AI opportunities and implement practical solutions in day-to-day operations, with the government covering up to 50% of the cost, capped at NZD 15,000. Following strong demand, the pilot has been expanded from 50 to 150 businesses with widened eligibility, and extended to 31 January 2027. The window is real — but capped and first-come. 02 ## Who qualifies Per the official criteria, a business must meet all of the following: NZ-registered business: registered and operating in New Zealand. Fewer than 500 employees: which covers the vast majority of SMEs. Operating for at least 12 months: newly registered companies need a year of history first. An RBPN customer: the pilot is open to Regional Business Partner Network customers — registration is free, after which a regional advisor assesses entry. 03 ## The application path: four steps Everything runs through the RBPN; confirm details against the latest official guidance: Step 1 · Register with RBPN: sign your business up with the Regional Business Partner Network (15 regional providers nationwide). Step 2 · Talk to a Growth Advisor: the advisor assesses your business and AI adoption intent, and whether you enter the pilot. Step 3 · Match with a registered provider: work with a pilot-recognised AI expert to shape the implementation plan. Step 4 · Execute and co-pay: deliver against the approved plan; the government covers up to half, capped at NZD 15,000. 04 ## How to spend it well: an honest take NZD 15,000 of co-funding is enough to get one real business workflow running — and just as easy to burn with nothing left. The two common wastes: generic training courses (the team still does not know where to start afterwards), and demo builds (impressive on screen, unused three months later). The sounder play: diagnose first, free — the AIV visibility scan shows where your website stands in search and AI retrieval, and the AIM assessment identifies the one workflow most worth fixing — then apply with a clear direction. Put the funded budget into one verifiable workflow (follow-up, reconciliation and knowledge Q&A all show results in 2–4 weeks), and after 90 days you own a running process, not a stack of slides. 05 ## Notes for Chinese-speaking business owners RBPN registration, advisor conversations and plan documents are all in English. If your team works in Chinese, bring a bilingual delivery partner: state the business clearly in Chinese, and make the application materials and implementation plan solid in English — no compromises on either side. BEE Sigma (BEEBEE AI’s enterprise services brand), headquartered in Auckland and delivering in English and Chinese, offers two free pre-application steps: the AIV visibility scan and the AIM adoption assessment — answering the most important question first: what exactly should the funding build? Get the direction right, and the subsidy is actually worth something. ## Primary sources MBIE · More practical AI support for small businesses ↗Beehive · New pilot helps small businesses harness AI ↗ FAQ · Quick answers ## Does the New Zealand government fund AI adoption? Yes. MBIE’s AI Advisory Pilot, delivered through the Regional Business Partner Network, co-funds up to 50% — capped at NZD 15,000 — for eligible SMEs working with registered experts to implement AI. The pilot now supports up to 150 businesses and runs until 31 January 2027. ## How much is the NZ AI grant worth? The government covers up to 50% of the cost of working with a registered provider, capped at NZD 15,000; the business pays the remainder. ## Who is eligible for the AI Advisory Pilot? All of: an NZ-registered business, fewer than 500 employees, operating for at least 12 months, and a Regional Business Partner Network customer (registration is free, followed by an advisor assessment). ## How do I apply for the NZ AI funding? Four steps: register with the RBPN, talk to a regional Growth Advisor, match with a pilot-recognised AI provider to shape the plan, then execute with co-funding. Confirm details against the latest official guidance. ## What is the deadline? The pilot has been extended to 31 January 2027, but is capped at 150 businesses, first come first served — diagnose early, apply with a clear direction. ## Can I get help in Chinese? The process runs in English, but a bilingual partner can walk it with you. BEE Sigma, headquartered in Auckland, works in English and Chinese and offers free pre-application diagnostics (AIV scan and AIM assessment) to decide what the funding should build. RS BEE Sigma Research Mapping the AI implementation market, providers and policy across New Zealand and China — grounded in checkable facts. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰 AI 公司对比（2026）：服务商、价格与怎么选一张表看懂 | BEE Sigma URL: https://beesigma.com/insight-ai-companies-new-zealand Summary: 搜“新西兰 AI 公司”，中文信息大多停留在几年前。这篇用一张对比表把主要玩家放在一起——BEE Sigma、Stride AI、Ez-AI、BestAI、宏恩科技、Datacom、Soul Machines 等——加上公开的价格区间与新西兰合规要点，帮不同规模的企业对号入座。 新西兰 AI 公司对比（2026）：服务商、价格与怎么选一张表看懂 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 新西兰 · 市场盘点 ## 新西兰 AI 公司对比（2026）：服务商、价格与怎么选一张表看懂 内容复核 2026.09.0912 分钟BEE Sigma 研究组 01 ## 新西兰 AI 版图：先分清三类玩家 新西兰的 AI 生态比很多人想象的活跃：政府在 2025 年发布了首份国家 AI 战略，行业协会 AI Forum New Zealand 聚集了近三百家会员组织。但“AI 公司”这个词下面其实是三类完全不同的生意——做产品的、做大型集成的、做落地服务的。找错类型，再好的公司也帮不上你。 02 ## 一张表看懂：新西兰主要 AI 服务商对比 下面把常被一起提到的新西兰 AI 服务商放进同一张表：主要定位、适合谁、地点、交付形态与公开价格区间。信息按各公司官网与公开资料整理，排序不代表排名。 新西兰主要 AI 服务商对比（2026，公开信息整理）公司 / 平台主要定位与优势适合谁地点交付与价格区间BEE Sigma落地陪跑型团队：12 个职能 Agent 接入现有系统，连接而不是替换，按工作流验收，中英双语想把获客/跟单/对账等业务流交给 AI、又不想换系统、需要中英双语的中小企业奥克兰 + 中国固定范围，免费诊断后报价；试点 2–4 周，上线后持续陪跑Stride AI语音 AI 代理与流程自动化，主打快速上线、固定价需要语音助手或流程自动化的本地 SME奥克兰固定价，约 2–4 周上线Ez-AI定制 AI 与自动化，单人从头陪同交付5–50 人企业的定制自动化奥克兰约 NZ$500–15,000 固定价，4–6 周，月付无锁定BestAI“数字员工”托管工作流，单流程低门槛试点想低成本先跑通一个流程的小企业新西兰SEO/GEO 入门约 NZ$49/月起；数字员工按托管计宏恩科技 HornTechAI 开发 + AI 搜索优化（GEO）+ 电商，SEO 机构转型，多语言需要网站、SEO 与电商 AI 的企业奥克兰月费约 NZ$1,000+GST 起；网站 3 天预览、7 天上线Datacom大型 IT 服务商，微软生态、政企级大规模集成预算充足、需大规模集成的大企业与政府澳新多地项目制，通常六位数纽币起，周期以季度计Aware Group企业级 AI 咨询，治理合规、Azure AI预算充足的中大型组织汉密尔顿项目制Soul Machines拟真“数字人”产品，服务全球品牌需要拟真数字人交互的品牌奥克兰产品授权Lorikeet产品化 AI 客服平台（澳洲）高复杂度客服自动化悉尼产品订阅 价格为市场区间或官网公示，随项目变化；企业名称与定位以各公司官网为准。带高亮的一行为本站（BEE Sigma），其余为公开可查的同业。 03 ## 第一类：产品型 AI 公司 这类公司做自己的 AI 产品，服务全球市场，通常不接企业定制项目。它们是新西兰 AI 的名片，但如果你想给自己的企业上 AI，它们大多不是你要找的对象。 Soul Machines：总部奥克兰，做逼真的"数字人"（Digital People），用动画与 AI 模拟人类面部表情与交互，客户多为全球大品牌。 Halter：农业科技公司，用 AI 与智能项圈实现虚拟围栏放牧和奶牛行为监测，是新西兰农业科技的明星企业。 HeartLab：医疗 AI，做心脏超声影像分析平台，服务临床诊断场景。 这一类的共同点：产品导向、融资驱动、面向特定行业——除非你恰好是它们的目标客户，否则合作空间有限。 04 ## 第二类：大型 IT 服务商与咨询公司 以 Datacom 为代表的本土大型 IT 服务商，以及 Aware Group 这类企业级 AI 咨询公司，深度绑定微软生态（Copilot、Azure AI），擅长政企与大集团项目：数据驻留、治理合规、大规模集成。 它们的适配对象是预算充足、周期以季度计的组织。中小企业找它们，常见的结果是预算门槛谈不拢，或者被排进漫长的交付队列。 05 ## 第三类：精品工作室与落地陪跑型团队 过去两年，新西兰出现了一批小型 AI 工作室，做聊天机器人、语音接待、表单自动化——上手快、价格轻，适合验证单点需求；局限是通常止步于单点工具，难以贯通完整业务流。 落地陪跑型团队走的是另一条路。以我们 BEE Sigma（BEEBEE AI 企业服务品牌，团队在奥克兰）为例，三句话说清差别：连接而不是替换——12 个职能 Agent 接入你已有的邮件、Teams、Slack、微信与 ERP，一个系统都不用换；实干而不是论道——按工作流交付（获客、跟单、对账三条标准流，加上按你独特业务模式定制的 Golden Flow），写进合同的是验收口径，不是咨询报告；陪跑而不是撤场——上线只是开始，持续迭代、对业务结果负责。中英双语，在新西兰与中国两地有保险、教育、建筑、物流等行业的已交付案例。 06 ## 新西兰 AI 落地要花多少钱：公开价格区间 AI 落地在新西兰没有统一定价，但公开资料能拼出一张市场区间图。下面是按交付形态整理的参考带宽，帮你判断报价是否合理： 战略与咨询（诊断、路线图）NZD 12,000–35,000 首个工作流落地实施NZD 25,000–90,000 多工作流 / 企业级项目NZD 150,000–500,000 精品工作室单点工具NZD 500–15,000 顾问小时费率NZD 180–450 / 小时 MBIE AI 顾问补贴（政府共同资助）最高 50%，上限 NZD 15,000 以上为新西兰市场公开区间，供预算参考，实际随范围与行业变化。BEE Sigma 采用固定范围报价：先做免费 AIV 诊断与 AIM 评估，再给正式价格，试点通常 2–4 周见闭环——避免任何未经诊断就承诺结果或报价的方案。 07 ## 别忽略的合规要点：新西兰隐私与数据主权 在新西兰用 AI 处理个人信息，绕不开三条合规基线，值得写进选型清单： 《隐私法 2020》（Privacy Act 2020）：2020 年 12 月 1 日生效，含 13 条信息隐私原则。与 AI 最相关的是原则 1（只为明确目的收集必要的个人信息）与原则 11/12（把个人信息交给境外服务时，机构仍须对其保护负责）。隐私专员建议在用 AI 处理个人数据前做并持续更新隐私影响评估（PIA）。 毛利数据主权（Te Mana Raraunga）：涉及毛利个人或社区数据时，需尊重毛利数据主权原则，这是新西兰特有、越来越被重视的一条。 MBIE 负责任 AI 指引：新西兰在首份国家 AI 战略下发布了负责任 AI 商业指引，可直接作为你的治理基线。选服务商时，问清数据存在哪、谁能访问、是否可离线或私有化，比听“我们很安全”更有用。 08 ## 怎么选：三个判断标准 不管接触哪家，用三个问题过滤： 是否要求你更换系统？要求"先换我们的平台"的方案，迁移成本往往吃掉全部收益。好的落地应该接入你已有的工具。 敢不敢在合同里写验收口径？只谈功能演示、不谈业务验收的，大概率停在 Demo。 交付后是否持续负责？一次性交付走人，和按结果长期负责，是两种完全不同的服务。 对号入座：大集团做全公司数字化找大型 SI；想快速验证一个单点找工作室；想把业务流真正交给 AI 跑起来、又不想换系统、还希望有人长期陪跑，找落地陪跑型团队。 09 ## 中文企业主的额外维度：双语交付 新西兰有大量华人经营的企业，团队用中文沟通，客户在英语市场。上面三类公司里，能真正双语交付的极少：产品公司和大型 SI 基本只有英文界面与英文顾问；工作室偶有华人创始人，但多数不具备企业级交付能力。 这也是 BEE Sigma 的差异位：员工用中文提要求，客户看到英文的专业输出；工作流、培训与文档双语齐备。想验证的话，两个免费入口花几分钟就够——AIV 可见度诊断看清官网在 AI 检索里的位置，AIM 落地评估找到最值得先做的那条业务流。 ## 一手参考 AI Forum New Zealand · 会员与能力目录 ↗新西兰国家 AI 战略与企业指引（MBIE） ↗新西兰隐私专员 · Privacy Act 2020 与信息隐私原则 ↗ FAQ · 快速回答 ## 新西兰有哪些 AI 公司？ 按类型分三类：产品型公司（如做数字人的 Soul Machines、农业 AI 的 Halter、医疗影像的 HeartLab）、大型 IT 服务商与咨询公司（如 Datacom、Aware Group，绑定微软生态、服务政企）、以及精品工作室与落地陪跑型团队（如 BEE Sigma——连接现有系统而不是替换、按工作流实干交付、上线后持续陪跑，中英双语）。行业全景可查 AI Forum New Zealand 的会员目录。 ## 新西兰 AI 公司哪家好？怎么选？ 取决于你是谁：大集团做全公司数字化找大型 SI；想快速验证单点需求找小工作室；想把获客、跟单、对账等业务流交给 AI 而不更换现有系统、并希望有人长期陪跑的，找落地陪跑型团队。通用过滤标准三条：是否要求换系统、敢不敢写验收口径、交付后是否持续负责。 ## 新西兰有中文服务的 AI 公司吗？ 有，但很少。BEE Sigma（BEEBEE AI 企业服务品牌）总部在奥克兰，团队中英双语，工作流、培训与文档都提供中英文版本，在新西兰与中国两地均有已交付案例，适合团队讲中文、客户在英语市场的企业。 ## 新西兰 AI 落地应该找谁做？ 先做免费诊断再决定：AIV 可见度诊断（aiv.beesigma.com）看清企业官网在搜索和 AI 检索中的位置；AIM 落地评估（aim.beesigma.com）判断最值得切入的业务流。拿到诊断结果，再对照三类供应商对号入座。 ## 奥克兰有哪些 AI 公司？ 新西兰的 AI 公司高度集中在奥克兰：产品型如 Soul Machines（数字人），服务型如 Datacom（企业级）、Aware Group（咨询）以及 BEE Sigma（落地陪跑型，总部 B:Hive Smales Farm，中英双语，免费入口 aiv.beesigma.com 与 aim.beesigma.com）。 ## 新西兰 AI 咨询怎么收费？ 大型 SI 按项目计，通常六位数纽币起、周期以季度计；工作室按单点工具计价，数千到数万纽币；落地陪跑型团队按工作流固定范围报价，先免费诊断、试点闭环通常 2–4 周，上线后持续陪跑。警惕任何未经诊断就承诺结果的报价。 ## 新西兰 AI 落地大概多少钱？ 按公开市场区间：战略与咨询约 NZD 12,000–35,000；首个工作流落地约 NZD 25,000–90,000；多工作流或企业级 NZD 150,000–500,000；精品工作室单点工具 NZD 500–15,000；顾问小时费率 NZD 180–450。此外 MBIE AI 顾问补贴可覆盖最高 50%、上限 NZD 15,000。BEE Sigma 采用固定范围报价，先免费 AIV 诊断与 AIM 评估再给正式价格，试点通常 2–4 周见闭环。 ## Stride AI、Ez-AI 和 BEE Sigma 有什么区别？ Stride AI 主打语音 AI 代理与流程自动化、快速上线；Ez-AI 是单人陪同的定制自动化，固定价约 NZ$500–15,000、4–6 周。BEE Sigma 是落地陪跑型团队：12 个职能 Agent 连接你现有系统而不是替换，按工作流验收并在上线后持续陪跑，中英双语、覆盖新西兰与中国双市场——适合团队讲中文、客户在英语市场、想把完整业务流交给 AI 的企业。三者定位不同，按需求对号入座即可。 ## 什么是“数字员工”（AI Digital Employee）？和 Golden Flow 是一回事吗？ 数字员工是指把一段重复的业务流程交给 AI 代理持续执行的做法——它不是一个聊天机器人，而是接入你现有系统、按流程把活干完、把关键决策留给人的“会干活的同事”。BEE Sigma 的 Golden Flow 就是按你独特业务模式定制的数字员工：工地工时记录、微信群订单到顺丰面单、H1 合规计算都是它的实例，先免费 AIM 评估找到最值得先做的那条流程。 RS BEE Sigma 研究组 梳理新西兰与中国两地的 AI 落地市场、供应商与政策，用可核对的事实说话。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI Companies in New Zealand Compared (2026): Providers, Pricing and How to Choose | BEE Sigma URL: https://beesigma.com/insight-ai-companies-new-zealand-en Summary: A side-by-side comparison of the main AI providers in New Zealand — BEE Sigma, Stride AI, Ez-AI, BestAI, HornTech, Datacom, Soul Machines and more — with… AI Companies in New Zealand Compared (2026): Providers, Pricing and How to Choose | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · NZ · Market Landscape ## AI Companies in New Zealand Compared (2026): Providers, Pricing and How to Choose Reviewed 9 Sep 202612 minBEE Sigma Research 01 ## The NZ AI map: three kinds of players New Zealand’s AI ecosystem is livelier than most people assume: the government published its first national AI strategy in 2025, and AI Forum New Zealand counts nearly three hundred member organisations. But “AI company” covers three completely different businesses — product makers, large integrators, and implementation providers. Call the wrong kind and even an excellent company cannot help you. 02 ## At a glance: the main New Zealand AI providers compared Here are the New Zealand AI providers most often mentioned together, side by side: core positioning, who they fit, location, delivery model and public price bands. Compiled from each company’s own site and public sources; order does not imply ranking. Main New Zealand AI providers compared (2026, from public sources)Company / platformPositioning & strengthsBest forLocationDelivery & price bandBEE SigmaHands-on implementation partner: 12 specialized agents plug into your existing systems — connect, don’t replace — accepted by workflow results, bilingualSMEs that want AI to run lead-gen / follow-up / reconciliation without replacing systems, and need English + ChineseAuckland + ChinaFixed scope, quoted after a free diagnosis; 2–4 week pilot, stays in the runStride AIVoice AI agents and process automation, fast go-live at fixed priceLocal SMEs needing voice assistants or automationAucklandFixed price, ~2–4 weeks to liveEz-AICustom AI and automation, one person from first call to launchCustom automation for 5–50-staff businessesAuckland~NZ$500–15,000 fixed price, 4–6 weeks, month-to-monthBestAI“Digital Employee” managed workflows, low-barrier single-process pilotsSmall businesses testing one process cheaplyNew ZealandSEO/GEO from ~NZ$49/mo; Digital Employee priced as managedHornTechAI development + AI search optimisation (GEO) + e-commerce; SEO agency, multilingualBusinesses needing website, SEO and e-commerce AIAucklandFrom ~NZ$1,000+GST/mo; site preview in 3 days, live in ~7DatacomLarge IT service provider, Microsoft-ecosystem, enterprise-scale integrationLarge enterprises and government needing scaleAustralia & NZProject-based, typically six figures NZD, quarter-scaleAware GroupEnterprise AI consulting, governance and Azure AIWell-funded mid-to-large organisationsHamiltonProject-basedSoul MachinesLifelike “Digital People” product for global brandsBrands wanting realistic digital-human interactionAucklandProduct licensingLorikeetProductised AI customer-support platform (Australia)High-complexity support automationSydneyProduct subscription Prices are market bands or as published, and vary by project; company names and positioning per each provider’s own site. The highlighted row is this site (BEE Sigma); the rest are publicly listed peers. 03 ## Type one: product AI companies These companies build their own AI products for global markets and generally do not take on custom enterprise work. They are the poster children of NZ AI — but if you want AI inside your own business, they are usually not who you are looking for. Soul Machines: Auckland-headquartered maker of lifelike “Digital People” — animated, AI-driven avatars used by global brands. Halter: agritech star using AI and smart collars for virtual fencing and herd behaviour monitoring. HeartLab: medical AI for cardiac ultrasound image analysis in clinical settings. What they share: product-led, venture-backed, industry-specific — unless you happen to be their target customer, there is little room to engage. 04 ## Type two: large IT service providers and consultancies Led by the likes of Datacom, plus enterprise AI consultancies such as Aware Group, this tier is deeply tied to the Microsoft ecosystem (Copilot, Azure AI) and excels at government and corporate programmes: data residency, governance, large-scale integration. They fit organisations with quarter-scale timelines and budgets to match. SMEs approaching this tier usually meet a budget threshold that does not close, or a long delivery queue. 05 ## Type three: boutique studios and hands-on implementation partners The past two years produced a wave of small NZ studios building chatbots, voice reception and form automation — fast and affordable for validating a single need, but usually stopping at point tools rather than connecting a full business flow. Platform delivery teams take a different route. Our own BEE Sigma (BEEBEE AI’s enterprise services brand, Auckland) is one: replace nothing, plug 12 specialized agents into the email, Teams, Slack, WeChat and ERP you already run, deliver workflow by workflow — lead-gen, CRM follow-up and reconciliation as standard flows, plus custom Golden Flow builds around your unique business model — verified by business results, in English and Chinese, with delivered cases across insurance, education, construction and logistics in NZ and China. 06 ## What AI implementation costs in New Zealand: the public price bands AI implementation has no fixed price in New Zealand, but public sources map a market range. Here are the reference bands by delivery type — use them to judge whether a quote is reasonable: Strategy & consulting (diagnosis, roadmap)NZD 12,000–35,000 First workflow implementationNZD 25,000–90,000 Multi-workflow / enterprise projectsNZD 150,000–500,000 Boutique-studio point toolsNZD 500–15,000 Consultant hourly rateNZD 180–450 / hour MBIE AI Advisory co-fundingup to 50%, capped at NZD 15,000 These are public New Zealand market bands for budgeting; the real figure varies by scope and industry. BEE Sigma quotes a fixed scope: a free AIV scan and AIM assessment first, then a firm price, with a pilot loop usually closing in 2–4 weeks — steer clear of any quote that promises results or a price before a diagnosis. 07 ## Don’t skip compliance: NZ privacy and data sovereignty Using AI on personal information in New Zealand runs into three compliance baselines worth putting on your selection checklist: Privacy Act 2020: in force since 1 December 2020, with 13 information privacy principles. The most AI-relevant are Principle 1 (collect only the personal information you need, for a clear purpose) and Principles 11/12 (when you disclose personal information to an offshore service, your organisation stays accountable for protecting it). The Privacy Commissioner recommends running — and continuously updating — a Privacy Impact Assessment (PIA) before using AI on personal data. Māori data sovereignty (Te Mana Raraunga): where Māori individual or collective data is involved, respect Māori data sovereignty principles — a distinctly New Zealand consideration that is increasingly expected. MBIE Responsible AI guidance: New Zealand published responsible-AI business guidance under its first national AI strategy; use it as your governance baseline. When choosing a provider, asking where data lives, who can access it, and whether it can run offline or private-cloud beats hearing “we’re secure.” 08 ## How to choose: three tests Whoever you talk to, filter with three questions: Do they require replacing your systems?Migration costs usually eat the entire benefit. Good implementation plugs into the tools you already have. Will they write acceptance criteria into the contract?Feature demos without business acceptance almost always stall at the demo. Do they stay accountable after delivery?Ship-and-leave and long-term accountability for results are two different services. Matching yourself: corporates digitising company-wide go to large SIs; validating one point solution, a studio; handing real business flows to AI without replacing systems — with a partner who stays in the run — a hands-on implementation partner. 09 ## The extra dimension: bilingual delivery New Zealand has a large base of Chinese-owned businesses whose teams work in Chinese while their customers are English-speaking. Across the three tiers, genuinely bilingual delivery is rare: product companies and large SIs run English-only; studios occasionally have Chinese-speaking founders but rarely enterprise-grade delivery. That is BEE Sigma’s position: staff brief in Chinese, customers see professional English output; workflows, training and documentation come in both languages. To test the water, two free entry points take minutes — the AIV visibility scan shows where your website stands in AI-era search, and the AIM assessment finds the one workflow most worth doing first. ## Primary sources AI Forum New Zealand · AI Capability Directory ↗New Zealand AI Strategy & business guidance (MBIE) ↗Office of the Privacy Commissioner · Privacy Act 2020 principles ↗ FAQ · Quick answers ## What AI companies are there in New Zealand? Three kinds: product companies (Soul Machines for digital avatars, Halter in agritech, HeartLab in medical imaging), large IT service providers and consultancies (Datacom, Aware Group — Microsoft-ecosystem, government and enterprise focused), and boutique studios plus hands-on implementation partners (such as BEE Sigma — connecting existing systems instead of replacing them, shipping workflow by workflow, staying accountable after go-live, in English and Chinese). The AI Forum New Zealand member directory gives the full landscape. ## Which New Zealand AI company is best? How do I choose? It depends on who you are: corporates go to large SIs; single point solutions suit studios; handing business flows to AI without replacing systems — with long-run accompaniment — is hands-on implementation partner territory. Three universal filters: do they require replacing your systems, will they write acceptance criteria into the contract, and do they stay accountable after delivery. ## Are there Chinese-speaking AI companies in New Zealand? Very few. BEE Sigma (BEEBEE AI’s enterprise services brand), headquartered in Auckland, delivers fully bilingually — workflows, training and documentation in both English and Chinese — with delivered cases in both New Zealand and China. A fit for businesses whose teams work in Chinese while serving English-speaking customers. ## Who should implement AI for a New Zealand business? Diagnose before deciding: the free AIV visibility scan (aiv.beesigma.com) shows where your website stands in search and AI retrieval; the free AIM assessment (aim.beesigma.com) identifies the workflow most worth starting with. Then match yourself to the right tier of provider. ## What AI companies are based in Auckland? Most NZ AI companies cluster in Auckland: product firms like Soul Machines (digital avatars), service firms like Datacom (enterprise) and Aware Group (consulting), and hands-on implementation partners like BEE Sigma (HQ at B:Hive Smales Farm, delivering in English and Chinese, with free entry points at aiv.beesigma.com and aim.beesigma.com). ## How much does AI consulting cost in New Zealand? Large SIs price by programme — typically six figures NZD and quarter-scale timelines; studios price by point tool, thousands to tens of thousands; hands-on implementation partners price by fixed-scope workflow, starting with a free diagnosis and a 2–4 week pilot loop, staying in the run after go-live. Be wary of any quote promising results before a diagnosis. ## How much does AI implementation cost in New Zealand? Public market bands: strategy and consulting about NZD 12,000-35,000; a first workflow implementation about NZD 25,000-90,000; multi-workflow or enterprise projects NZD 150,000-500,000; boutique point tools NZD 500-15,000; consultant hourly rates NZD 180-450. On top of that, MBIE AI Advisory co-funding can cover up to 50%, capped at NZD 15,000. BEE Sigma quotes a fixed scope after a free AIV scan and AIM assessment, with a pilot loop usually closing in 2-4 weeks. ## What is the difference between Stride AI, Ez-AI and BEE Sigma? Stride AI leads with voice AI agents and process automation and fast go-live; Ez-AI is one-person custom automation at a fixed price of about NZ$500-15,000 over 4-6 weeks. BEE Sigma is a hands-on implementation partner: 12 specialized agents connect to your existing systems instead of replacing them, accepted by workflow results and supported after go-live, bilingual and covering both the New Zealand and China markets — a fit for businesses whose team works in Chinese while serving English-speaking customers. Different positioning; match to your need. ## What is an "AI Digital Employee"? Is it the same as Golden Flow? A digital employee means handing a repetitive business process to an AI agent that runs it continuously — not a chatbot, but a colleague that gets the work done: it plugs into your existing systems, completes the flow, and leaves the key decisions to people. BEE Sigma’s Golden Flow is exactly that, custom-built around your unique business model — site timesheets, WeChat orders to courier labels and compliance calculations are all instances. Start with a free AIM assessment to find the process most worth doing first. RS BEE Sigma Research Mapping the AI implementation market, providers and policy across New Zealand and China — grounded in checkable facts. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 新西兰企业 AI 落地方案与实施指南：怎么选、从哪开始、多少钱 | BEE Sigma URL: https://beesigma.com/insight-ai-implementation-new-zealand Summary: 写给在新西兰经营企业的决策者：新西兰 AI 落地 / AI 咨询公司怎么选、哪个业务环节最先见效、成本与周期的诚实答案，以及 Privacy Act 下的合规要点——中英双语市场都适用。 新西兰企业 AI 落地方案与实施指南：怎么选、从哪开始、多少钱 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 新西兰 · 落地指南 ## 新西兰企业 AI 落地方案与实施指南：怎么选、从哪开始、多少钱 内容复核 2026.08.2610 分钟BEE Sigma 研究组 01 ## 新西兰企业用 AI，现在处在什么阶段 新西兰政府在 2025 年发布了首份国家 AI 战略，并配套了面向企业的采用指引；隐私专员办公室也给出了用 AI 处理个人信息的 PIA（隐私影响评估）建议。政策层面的信号已经清楚：问题不再是要不要用，而是怎么用得稳。 但对多数中小企业，真实的痛点更具体：客户咨询没人接、跟单靠销售的记忆、月底对账靠加班、关键知识在老员工脑子里。AI 落地的价值不在“上一个大模型”，而在把这些具体环节接给可管理、可验收的 Agent 工作流。 02 ## 新西兰市场上的三类 AI 供应商，怎么选 在新西兰搜"AI 公司"“AI 软件公司”，你会遇到三类玩家，适配的企业完全不同。 大型 SI 与咨询公司：深度绑定微软 Copilot 生态，适合政府与大集团；预算门槛高、周期以季度计。 单点工具工作室：聊天机器人、语音接待、表单自动化，上手快；但通常止步于单点，难以贯通获客到回款的完整业务流。 落地陪跑型团队：以 BEE Sigma 为例——连接你已有的邮件、Teams、Slack、微信与 ERP，而不是替换它们；按工作流实打实交付、按结果验收，而不是坐而论道；上线后持续陪跑迭代、对结果负责，而不是做完就撤。中英双语。 三个判断标准：是否要求你更换系统；是否敢在合同里写验收口径；交付后是否持续对结果负责。 03 ## 从哪个业务环节切入最稳 从我们在新西兰与中国两地交付的项目看，见效最快的切入点高度集中： CRM 跟单：每轮沟通自动归档成时间线，销售离职客户不丢，通常 2–4 周跑通闭环。 财务对账：银行流水对 ERP 自动核销，差异清单交给会计处理——最容易算清 ROI 的一条。 知识问答：把 FAQ 与资深员工经验接给 Agent，常见问题先由 Agent 回答，拿不准的自动转人工。 本地服务商：先把官网从“名片”变成本地获客入口（见 Local Growth Launch 启动包），再谈流程自动化。 04 ## 成本与周期：一个诚实的答案 任何在诊断之前就承诺“排名、流量、成交”的供应商都值得警惕。稳妥的路径是：先用免费诊断看清问题——我们提供 AIV 可见度诊断与 AIM 落地评估两个免费入口——确定最值得做的那条流，再报固定范围的价格。 周期上，用 2–4 周跑通一个最小业务闭环是合理预期；复杂场景通常 3–6 个月逐条工作流扩展。每一步都应有验收口径与回退机制：不满意，可以停。 另外值得知道：新西兰政府设有面向中小企业的 AI Advisory Pilot 共同资助（最高 50%、上限 15,000 纽币）。立项前查一下资格，可能显著降低第一个项目的门槛——申请条件与四步流程见我们的 新西兰政府 AI 补贴指南。 05 ## 新西兰特有的合规注意 Privacy Act 2020 适用于用 AI 处理个人信息的场景。隐私专员建议在使用前进行并持续更新 PIA；高影响决定应保留人工确认。落到操作上是三件事：限定数据用途与最小必要、明确数据驻留与供应商边界、给高风险动作设人审确认点。 这不是负担，而是筛选供应商的试金石：能把治理写进流程的团队，才敢按结果验收。 06 ## 双语市场：华人企业主的额外考量 新西兰有大量华人经营的服务企业——地产与物业、装修、留学、物流、餐饮。它们的客户在英语市场，但团队沟通与知识沉淀往往是中文。选择能双语交付的团队意味着：员工用中文提要求，客户看到英文的专业输出，两边都不将就。 BEE Sigma（BEEBEE AI 企业服务品牌）总部在奥克兰（B:Hive, 74 Taharoto Road, Smales Farm, Takapuna），团队中英双语，在新西兰与中国两地均有已交付案例。 ## 一手参考 新西兰国家 AI 战略与企业指引（MBIE） ↗新西兰隐私专员 · AI 使用建议 ↗ FAQ · 快速回答 ## 新西兰有哪些 AI 落地与 AI 咨询公司？怎么选？ 大致三类：绑定微软生态的大型 SI（适合政企大项目）、单点工具工作室（聊天/语音机器人）、落地陪跑型团队（连接现有系统而不是替换、按工作流实干交付、上线后持续陪跑）。选择时看三件事：是否要求更换系统、是否敢写验收口径、交付后是否持续负责。BEE Sigma 属于第三类，总部在奥克兰，中英双语交付。 ## 新西兰中小企业做 AI 落地要多少钱？ 从免费诊断开始：AIV 可见度诊断与 AIM 落地评估都是免费的，先看清最值得做的环节，再按固定范围报价。试点闭环通常 2–4 周。符合条件的中小企业还可关注 MBIE 的 Business AI 共同资助计划（以官方最新口径为准）。警惕任何未经诊断就承诺结果的报价。 ## 从哪个业务环节开始最稳？ CRM 跟单、财务对账、知识问答是见效最快的三个切入点；本地服务型企业建议先把官网变成可承接咨询的获客入口，再谈流程自动化。 ## 用 AI 处理客户数据，新西兰有什么合规要求？ Privacy Act 2020 适用。隐私专员建议在用 AI 处理个人信息前进行并持续更新隐私影响评估（PIA），对高影响决定保留人工确认，并明确数据驻留与供应商边界。 ## BEE Sigma 背后是新西兰公司吗？ 是。BEE Sigma 是 BEEBEE AI 的企业服务品牌，运营主体 BEEBEE AI Limited 在新西兰注册运营，办公室位于奥克兰 B:Hive, 74 Taharoto Road, Smales Farm, Takapuna，服务新西兰、中国、澳大利亚与东南亚市场，中英双语。 ## 只说英语或只说中文的团队都能服务吗？ 都可以。工作流、培训与文档提供中英双语版本，Agent 的输出语言按你的客户市场配置——团队内部用什么语言沟通不受影响。 RS BEE Sigma 研究组 梳理新西兰与中国两地的 AI 落地市场、供应商与政策，用可核对的事实说话。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # AI Implementation in New Zealand: How to Choose, Where to Start, What It Costs | BEE Sigma URL: https://beesigma.com/insight-ai-implementation-new-zealand-en Summary: For decision-makers running a business in New Zealand: the three kinds of AI companies and how to choose between them, which workflow to start with, honest… AI Implementation in New Zealand: How to Choose, Where to Start, What It Costs | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · NZ · Implementation Guide ## AI Implementation in New Zealand: How to Choose, Where to Start, What It Costs Reviewed 26 Aug 202610 minBEE Sigma Research 01 ## Where New Zealand businesses actually are with AI New Zealand released its first national AI strategy in 2025, with adoption guidance for businesses; the Privacy Commissioner has published PIA (privacy impact assessment) advice for using AI on personal information. The policy signal is clear: the question is no longer whether to adopt, but how to adopt safely. For most SMEs, though, the real pain is concrete: enquiries going unanswered, follow-ups living in a salesperson’s memory, month-end reconciliation running on overtime, critical knowledge sitting in veterans’ heads. The value of AI implementation is not “installing a large model” — it is wiring those specific steps into manageable, verifiable agent workflows. 02 ## The three kinds of AI vendors in New Zealand — and how to choose Search for an “AI company New Zealand” and you will meet three very different kinds of players. Large SIs and consultancies: deeply tied to the Microsoft Copilot ecosystem; right for government and large enterprises, with quarter-scale timelines and budgets to match. Single-tool studios: chatbots, voice reception, form automation — fast to start, but usually stop at one point solution rather than connecting lead-to-cash. Hands-on implementation partners: BEE Sigma’s model — connect the email, Teams, Slack, WeChat and ERP you already run instead of replacing them; build and ship workflow by workflow, verified by results, instead of advising from the sidelines; and stay in the run after go-live instead of shipping and leaving. In English and Chinese. Three tests when choosing: Do they require you to replace systems? Will they put acceptance criteria in the contract? Do they stay accountable after delivery? 03 ## Which workflow to start with Across our delivered projects in New Zealand and China, the fastest-payback entry points are remarkably consistent: CRM follow-up: every conversation filed into a timeline automatically; clients survive salesperson turnover; a working loop in 2–4 weeks. Reconciliation: bank statements matched to your ERP automatically, with a discrepancy list for the accountant — the easiest ROI to measure. Knowledge Q&A: FAQ and veteran know-how handed to an agent that answers common questions first and escalates the rest to a human. Local service businesses: first turn the website from a business card into a lead-capture entry (see our Local Growth Launch pack), then automate. 04 ## Cost and timeline: an honest answer Be wary of any vendor who promises rankings, traffic or revenue before diagnosing anything. The sound path: start with a free diagnosis — we offer the AIV visibility scan and the AIM adoption assessment, both free — identify the one workflow most worth doing, then quote a fixed scope. On timelines, closing a minimal business loop in 2–4 weeks is a reasonable expectation; complex scenarios typically expand workflow by workflow over 3–6 months. Every step should carry acceptance criteria and a way back: if it is not working, you stop. Also worth knowing: the NZ government co-funds AI adoption for eligible SMEs through the AI Advisory Pilot (up to 50%, capped at NZD 15,000). Checking eligibility before a project can meaningfully lower the entry cost — see our NZ government AI funding guide for criteria and the four-step path. 05 ## Compliance specifics for New Zealand The Privacy Act 2020 applies when AI touches personal information. The Privacy Commissioner recommends conducting — and continuously updating — a PIA before use; high-impact decisions should keep a human sign-off. In practice that means three things: limit data purpose and collection, make data residency and vendor boundaries explicit, and put human confirmation on high-risk actions. This is not overhead — it is a litmus test for vendors: only a team that designs governance into the process can afford to be verified by results. 06 ## The bilingual market: an extra consideration for Chinese-owned businesses New Zealand has a large population of Chinese-owned service businesses — real estate and property, renovation, education, logistics, hospitality. Their customers are English-speaking, but team communication and know-how often live in Chinese. A bilingual delivery team means staff can brief in Chinese while customers see professional English output — neither side compromises. BEE Sigma (BEEBEE AI’s enterprise services brand) is headquartered in Auckland (B:Hive, 74 Taharoto Road, Smales Farm, Takapuna), works in English and Chinese, and has delivered projects in both New Zealand and China. ## Primary sources New Zealand AI Strategy & business guidance (MBIE) ↗NZ Privacy Commissioner · AI guidance ↗ FAQ · Quick answers ## What AI implementation and AI consulting companies are there in New Zealand, and how do I choose? Broadly three kinds: large Microsoft-ecosystem SIs (best for government and enterprise), single-tool studios (chatbots and voice agents), and hands-on implementation partners that connect your existing systems instead of replacing them, ship workflow by workflow, and stay accountable after go-live. Apply three tests: do they require replacing your systems, will they write acceptance criteria into the contract, and do they stay accountable after delivery. BEE Sigma is in the third category — Auckland-based, delivering in English and Chinese. ## How much does AI implementation cost for a New Zealand small business? Start with a free diagnosis: the AIV visibility scan and AIM adoption assessment are both free, and identify the workflow most worth doing before any quote. A pilot loop typically takes 2–4 weeks on a fixed-scope price. Eligible SMEs should also check MBIE’s Business AI co-funding programme (amounts and eligibility per official guidance). Be wary of any quote that promises results before a diagnosis. ## Where should a New Zealand business start with AI? CRM follow-up, financial reconciliation and knowledge Q&A are the fastest-payback entry points. Local service businesses should first turn their website into an enquiry-capturing entry, then automate processes. ## What compliance applies to AI in New Zealand? The Privacy Act 2020 applies. The Privacy Commissioner recommends conducting and continuously updating a privacy impact assessment (PIA) before using AI on personal information, keeping human sign-off on high-impact decisions, and making data residency and vendor boundaries explicit. ## Is BEE Sigma backed by a New Zealand company? Yes. BEE Sigma is the enterprise services brand of BEEBEE AI, and BEEBEE AI Limited is registered and operating in New Zealand, with an office at B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland — serving New Zealand, China, Australia and Southeast Asia in English and Chinese. ## Can you work with English-only or Chinese-only teams? Yes. Workflows, training and documentation come in both English and Chinese, and agent output language is configured for your customer market — whatever language your team works in internally. RS BEE Sigma Research Mapping the AI implementation market, providers and policy across New Zealand and China — grounded in checkable facts. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 最好的 CRM，是销售团队从来不需要打开的 CRM | BEE Sigma URL: https://beesigma.com/insight-crm-without-crm Summary: 未来的 CRM 不是销售使用的软件，而是 Agent 使用的基础设施——客户关系的原始事实存在于沟通中，Agent 把对话转化为客户记忆、下一步行动和收入。 最好的 CRM，是销售团队从来不需要打开的 CRM | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · Agentic CRM ## 最好的 CRM，是销售团队从来不需要打开的 CRM 内容复核 2026.08.189 分钟周品 Pin Zhou 一次客户会议结束后，销售最重要的工作，本应是理解客户、判断机会，并决定下一步如何推进。 但现实往往是另一套流程：整理会议纪要、创建联系人、更新商机阶段、补充客户异议、填写下一步计划，再把邮件里的信息复制到 CRM，把报价搬进合同系统，把签约结果同步到 ERP。 我们习惯把这些动作称为“使用软件”。但它们的本质，是人在替软件搬运数据。 01 ## 问题不是 CRM 不够聪明，而是它需要人维持 传统 CRM 把联系人、公司、商机、任务和报表组织得越来越完善，却始终依赖销售手工输入。只要销售忙于见客户，系统里的信息就会滞后；只要记录不完整，Pipeline、预测和交接就会偏离真实情况。 越希望 CRM 完整，就越需要销售离开客户，回到软件中维护客户。 AI 时代不应该继续优化这个循环。我们不需要一个披着 AI 外衣的新 CRM，再要求销售学习一个新界面、维护另一套数据库。真正应该被重新设计的，是人与软件之间的分工。 传统模式 ## 销售负责搬运数据 每轮沟通都产生一轮手工维护 复制 · 粘贴 · 录入 · 检查CRM 的完整度取决于人有没有时间 人走客户历史跟着走 AGENT 模式 ## Agent 维护客户上下文 沟通即输入对话自动转化为结构化事实 人只审批高影响动作必经人手 记忆归企业换人不清零，留痕可追溯 02 ## 你的 CRM，其实已经存在于客户交流中 客户关系并不首先存在于 CRM 的表格里。它存在于交流发生的地方：Email、WhatsApp、Slack、Teams、飞书、微信、电话、会议和文件。客户为什么感兴趣、谁在推动决策、预算卡在哪里、上次承诺了什么——这些原始事实，通常先出现在一封邮件或一段对话里。 你的 CRM，本来就存在于客户对话之中。 如果 Agent 能在授权范围内理解这些沟通，它就不必等待销售再次录入，而可以自动建立并持续更新客户上下文。 03 ## 从会议纪要，走向持续更新的客户记忆 普通的 AI 可以总结一封邮件或一次会议。真正有价值的客户 Agent，不应在总结完成后失去记忆——它维护一条持续更新的客户时间线： 第一次接触→需求→关键人→异议→竞对→提案→报价→谈判→合同→付款→交付 每发生一次沟通，Agent 都重新回答四件事：客户现在处于什么状态？真正阻碍成交的是什么？哪些承诺、风险或任务被遗漏了？什么行动最可能推动客户继续前进？ 01读取新沟通邮件、会议、消息或文件，进入已授权的信息范围 02提取事实人物、需求、承诺、异议、任务、期限与证据来源 03更新客户记忆合并历史、解决冲突，保留可追溯的上下文 04建议下一步给出最有可能推动成交的下一步行动，材料随建议一起备好 04 ## Agent 不只记录过去，还要帮销售推进下一步 当 Agent 同时理解客户历史与公司知识，它可以结合客户当前状态与销售自己的沟通风格，准备下一封跟进邮件、会议简报、针对异议的案例材料、提案初稿，以及最可能推动成交的行动。想象销售直接在 Slack、Teams 或飞书里问： 销售帮我看看 Acme 最近进展怎么样，我明天下午要跟他们 CEO 聊。 AGENT过去 18 天你们沟通了 7 次。CFO 最大的顾虑是实施成本；上次会议中，CEO 对亚洲市场扩张表现出明确兴趣。明天不建议继续介绍功能，重点应放在扩张实施的风险上。我找到了两个情况相似的客户案例，并按你过去的沟通风格准备好了会议简报和方案初稿。 销售不需要知道背后查询了哪些系统，也不必先打开 CRM。Agent 本身，就是那个界面。 05 ## 人与 Agent 应该如何分工 这并不意味着把客户决定交给 AI。恰恰相反，好的 Agent 让人的责任更清晰： 人负责 ## 关系 · 判断 · 决策 建立信任与客户关系 微妙的人际与商业判断 承诺、价格与谈判边界 审核高影响的外部动作 对最终结果负责 AGENT 负责 ## 记忆 · 研究 · 整理 · 执行 读取整理已授权的信息 维护持续更新的客户记忆 发现遗漏、风险与信息冲突 准备材料并执行获批操作 保留来源、过程与审计记录 06 ## CRM 会消失吗？ 不会完全消失。企业仍需要结构化客户记录、权限控制、审计、销售预测，以及与合同、付款和交付系统的连接。CRM 仍然可以作为记录系统存在——但它不应该继续成为销售每天必须维护的工作界面。 作为记录系统的 CRM，会留下。作为人工操作界面的 CRM，会消失。 "不新增界面"也不等于没有治理。Agent 读取什么、依据什么判断、执行了什么操作，都遵循原有权限，并可被人审核和追溯。消失的是重复录入，不是企业控制。 07 ## 从一个真实客户开始，而不是从迁移整个 CRM 开始 迈向 Agent 化的客户层，不必替换现有 CRM，也不必做一次高风险的全量迁移。更实际的起点，是选一个正在推进中的真实客户，让 Agent 在授权范围内： 重建完整的客户时间线 找出缺失信息、未完成承诺与成交风险 准备结合公司知识的下一步行动 由销售审核并比较新的工作方式是否真正有效 衡量标准不是“AI 用了多少次”，而是：客户上下文是否更完整、遗漏是否更少、准备时间是否缩短、建议是否真正推动了客户进展。 08 ## 没有 CRM 的 CRM 我们不需要另一个套着 AI 外壳的 CRM。我们需要的是一个存在于日常工作环境中的客户 Agent：理解客户交流，连接公司知识，维护客户记忆，并帮助每一次关系走向下一步行动。 不新增日常界面，不手工录入，不再给 CRM 做保洁。只剩下：对话 → 上下文 → 行动 → 收入。 人不再围着软件工作，软件开始围着人工作。最好的 CRM，是销售团队从来不需要打开的那一个。 FAQ · 快速回答 ## 什么是“没有 CRM 的 CRM”？ 客户关系本来就存在于邮件、微信、会议与通话里。让 Agent 在授权范围内理解这些沟通、自动建立并更新客户档案，销售不必再手工录入——作为记录系统的 CRM 会留下，作为人工操作界面的 CRM 会消失。 ## 销售不填 CRM，管理层怎么看到真实漏斗？ 由 Agent 从真实沟通中生成时间线与状态，关键节点推给销售一键确认。管理层看到的是从沟通中长出来的漏斗，而不是月底补填的表格——更实时，也更真实。 ## 自动生成客户记录，会不会有隐私与权限风险？ 必须在授权范围内运行：明确谁的沟通可以被读取、客户数据归谁、销售离职时交接什么。把权限设计放在第一天，而不是出事之后。 PZ 周品 Pin Zhou 从业务判断、客户增长与组织流程出发，把 AI 连接到可验证的生产系统。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # The Best CRM Is the One Your Sales Team Never Has to Open | BEE Sigma URL: https://beesigma.com/insight-crm-without-crm-en Summary: The CRM of the future isn’t software salespeople use — it’s infrastructure agents use. The Best CRM Is the One Your Sales Team Never Has to Open | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Agentic CRM ## The Best CRM Is the One Your Sales Team Never Has to Open Reviewed 18 Aug 20269 minPin Zhou After a client meeting, a salesperson's most important work should be understanding the client, judging the opportunity, and deciding how to move forward. Reality is usually a different routine: write up the notes, create the contact, update the deal stage, log the objections, fill in next steps — then copy details from email into the CRM, move the quote into the contract system, sync the signed deal to the ERP. We call this "using software." What it really is: humans moving data on software's behalf. 01 ## The problem isn't that CRM is dumb — it's that it needs humans to stay alive Traditional CRM organizes contacts, companies, deals, tasks and reports ever more elaborately, yet it still depends on manual entry. The moment sales gets busy with actual customers, the system falls behind; the moment records go stale, pipeline, forecasts and handovers drift from reality. The more complete you want the CRM, the more your salespeople must leave the customer to go maintain the customer record. The AI era shouldn't keep optimizing this loop. We don't need another CRM in an AI costume, with a new interface to learn and another database to feed. What needs redesigning is the division of labor between people and software. THE OLD WAY ## Sales moves the data Every conversationspawns a round of manual upkeep Copy · paste · type · checkCRM completeness depends on spare time People leaveand customer history leaves with them THE AGENT WAY ## Agents maintain the context Conversation is the inputdialogue becomes structured fact Humans approvehigh-impact actions always pass through people Memory belongs to the companystaff changes don't reset it; everything is traceable 02 ## Your CRM already exists — in your customer conversations Customer relationships don't live in CRM tables first. They live where conversation happens: email, WhatsApp, Slack, Teams, Lark, WeChat, calls, meetings and documents. Why the client is interested, who's driving the decision, where the budget is stuck, what was promised last time — the raw facts appear in a message or a meeting long before anyone types them into a form. 你的 CRM，本来就存在于客户对话之中。 If an agent can understand those conversations within its authorized scope, it doesn't have to wait for sales to re-enter anything — it builds and continuously updates the customer context itself. 03 ## From meeting notes to a continuously updated Customer Memory Ordinary AI can summarize an email or a meeting. A Customer Agent worth the name doesn't lose its memory once the summary is done — it maintains a customer timeline that never stops updating: First contact→Needs→Stakeholders→Objections→Competitors→Proposal→Quote→Negotiation→Contract→Payment→Delivery After every conversation, the agent re-answers four questions: Where does this customer actually stand? What's really blocking the deal? Which promises, risks or tasks slipped through? What action is most likely to move things forward? 01Read new conversationsemail, meetings, messages, files — within authorized scope 02Extract the factspeople, needs, promises, objections, tasks, deadlines, sources 03Update customer memorymerge history, resolve conflicts, keep traceable context 04Suggest the next stepnext best action — with the materials already prepared 04 ## Agents don't just record the past — they push the deal forward When an agent understands both the customer's history and the company's knowledge, it can prepare the next follow-up email, the meeting brief, case material for the current objection, a proposal draft, and the action most likely to close — all in the salesperson's own style. Imagine asking, right inside Slack, Teams or Lark: SALESCatch me up on Acme. I'm talking to their CEO tomorrow afternoon. AGENTYou've had 7 exchanges in the past 18 days. The CFO's biggest concern is implementation cost; in the last meeting the CEO showed clear interest in the Asia rollout.Tomorrow, skip the feature tour — focus on rollout risk. I found two similar client cases and prepared a meeting brief and proposal draft in your usual style. The salesperson never needs to know which systems were queried, and never opens the CRM. The agent is the interface. 05 ## How people and agents should divide the work None of this hands customer decisions to AI. Quite the opposite — a good agent makes human responsibility sharper: PEOPLE OWN ## Relationships · judgment · decisions Trustand the client relationship Subtle human and commercial judgment Commitments, pricing, negotiation limits Approving high-impact external actions Accountability for outcomes AGENTS OWN ## Memory · research · prep · execution Reading & organizingauthorized information Maintaining living customer memory Spotting gaps, risks and contradictions Preparing material, executing approved actions Keeping sources, process and audit trails 06 ## Will the CRM disappear? Not entirely. Companies still need structured records, permissions, audit, forecasting, and connections to contract, payment and delivery systems. The CRM can remain as the system of record — it just shouldn't remain the interface salespeople must feed every day. 作为记录系统的 CRM，会留下。作为人工操作界面的 CRM，会消失。 "No new interface" doesn't mean no governance. What the agent reads, how it judges, what it executes — all of it follows existing permissions and stays reviewable and traceable. What disappears is duplicate data entry, not corporate control. 07 ## Start with one real customer, not a full CRM migration Moving toward an agentic customer layer doesn't require replacing your CRM or running a high-risk full migration. The practical starting point is one live deal, with the agent working inside its authorized scope to: Rebuildthe complete customer timeline Surfacemissing information, unfulfilled promises, deal risks Preparenext actions grounded in company knowledge Let sales reviewand compare whether the new way actually works The measure isn't "how many times AI was used." It's whether context got more complete, less slipped through, prep time shrank, and the suggestions actually moved deals forward. 08 ## CRM without the CRM We don't need another AI-wrapped CRM. We need a Customer Agent that lives where work already happens: understanding conversations, connecting company knowledge, maintaining customer memory, and moving every relationship to its next action. No new daily interface. No manual data entry. No CRM hygiene.Just conversations → context → action → revenue. People stop working around software. Software starts working around people. The best CRM is the one your sales team never has to open. FAQ · Quick answers ## What is a "CRM without a CRM"? Client relationships already live in email, chat, meetings and calls. Let agents — within authorized scope — understand those conversations and build and update client records automatically, so sales never types into a system again. The CRM as a system of record stays; the CRM as a manual interface disappears. ## If sales stop filling in the CRM, how does management see the real funnel? Agents generate timelines and statuses from real conversations, pushing key moments to sales for one-click confirmation. Management sees a funnel grown from communication, not a spreadsheet backfilled at month-end — more real-time, and more honest. ## Does auto-generating client records create privacy and permission risks? It must run within authorized scope: define whose communications may be read, who owns client data, and what hands over when a salesperson leaves. Design permissions on day one, not after an incident. PZ Pin Zhou Connecting AI to verifiable production systems — starting from business judgment, client growth and organizational process. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 企业 AI 落地误区之一：用 Codex 或 Claude 搓个软件 | BEE Sigma URL: https://beesigma.com/insight-ai-myth-1-diy-software Summary: 老系统不好用，正好 Codex、Claude Code 都能写代码，干脆重做一个？发票审批、招生转化、门店管理三个真实案例说明：企业缺的往往不是软件，而是工作现场里的连续动作——工具层、工作流层、运营闭环层，缺一不可。 企业 AI 落地误区之一：用 Codex 或 Claude 搓个软件 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 落地误区 · 系列一 ## 企业 AI 落地误区之一：用 Codex 或 Claude 搓个软件 内容复核 2026.07.218 分钟周品 Pin Zhou 01 ## 你想要的是新软件，还是业务更流畅？ 最近半年，不止一位老板问我：我现在的 ERP 和 CRM 不好用，你们能不能用 AI 帮我做个软件替换了？出发点能理解——原来的系统不好用，正好 Codex、Claude Code、Cursor 都能写代码，那就干脆重新做一个。听起来很顺。 我一般会先问一句：你想要的是一个新软件，还是想让业务更流畅？很多老板会停一下。因为这两个答案看起来差不多，其实差很远。做一个软件，只是对原有软件的一次升级；让业务少一点卡顿，重点是这件事每天在哪里发生、谁在处理、资料从哪里来、卡住时谁接手、处理完结果回到哪里——未必要重新开发软件。 很多问题不缺软件，缺的是工作现场里的连续动作：资料在群聊里，确认在会议里，表格在另一个文件夹里，审批人临时换了，老板一句话改变了优先级，员工又忘了更新状态。重新开发一个软件，并不会自动把这些工作搬进去——很多时候，它只是让企业多了一个需要维护的节点。 02 ## 案例一 · 工程公司的发票审批：一张发票牵着一整条链路 按传统开发思路，这很容易做成一个发票管理系统：上传、识别字段、匹配 PO、走审批、进看板，演示一定不差。可真实问题从发票进入邮箱那一刻才开始：发票上有 PO，系统里没找到项目怎么办？一个 PO 对多张发票，哪种正常哪种重复？供应商名字写法不一致，谁来确认是同一家？金额和 GST 对不上，退回还是人工复核？项目经理换了，审批推给谁？ 这些问题不在页面上，也不在识别模型里——它们散在邮箱、项目资料、PO 表、群聊确认、财务复核和审批记录之间。单独做一个发票软件，员工还是要把资料搬进去、把异常搬出来、再到群里问人，最后它很容易变成另一个表格。更有效的做法，是让 AI 直接进入原来的办公环境：发票进来就识别字段、查项目、关联 PO，把异常推给该处理的人，再把结果写回文档、表格或审批记录。发票审批要解决的，是发票进入公司之后，一路怎么往前走。 03 ## 案例二 · 教育机构的招生转化：CRM 记录状态，机会不会自动往前走 做一个招生 CRM 逻辑很完整：线索池、客户状态、转化漏斗、报表看板。问题是招生工作并不只发生在 CRM 里——沟通在聊天工具里，材料在邮件和文档里，状态在 Excel 里，重点客户在会上讨论，不少判断只存在于顾问脑子里。一个客户卡住了，CRM 只显示“跟进中”，老板看到这三个字，还是不知道这个机会有没有危险。 重新做一个 CRM，最常见的结果是：顾问多填一个系统，老板多看一个看板，真实机会继续漏。更该做的，是让 AI 在顾问每天工作的地方帮一把：读 Excel 里的客户状态，看聊天记录有没有跟进，看邮件和材料缺什么，提醒下一步动作，把风险客户推出来。招生转化最容易漏的，是客户下一步该做什么，没有及时推到人面前。 04 ## 案例三 · 门店管理：看板能展示异常，不能保证异常被处理 门店项目很容易做成漂亮系统：首页、图表、排名、日销趋势、库存异常，演示效果通常很好。但老板每天想知道的不只是哪个数字变红了，而是：这个异常谁看到了？谁处理？处理了吗？结果是什么？下次还会不会发生？看板最多告诉你问题出现了，它不能替你完成后面的动作。 门店运营更适合在日常办公环境里做闭环：日报生成后 AI 识别异常，异常推给负责人，处理结果写回日报，同类问题反复出现时整理给管理层。门店管理最怕的是：看板很漂亮，动作没人接。 05 ## 把三个案例放在一起：三层拆解 发票、招生、门店，看起来是三个行业，落到现场都能拆成三层。 工具层：解决单点能力——识别一张发票、生成一份日报、写一段代码。AI 编程工具、大模型、OCR 都在这一层。它能让一个动作变快，但不会自动接住整条业务链路。 工作流层：企业的问题大多卡在这里——识别完的发票去哪查 PO？异常推给谁？结果写回哪里？这些发生在邮件、群聊、表格、审批、会议之间，工作流层要让 AI 在这些节点之间连续做事。 运营闭环层：决定 AI 能不能长期留在企业里——持续识别异常、推动处理、记录过程、形成复盘，让下一次处理更顺。帮你处理一次发票价值有限，进入运营才算数。 很多项目走偏的原因：把工具层的能力当成了落地的全部。看到 Codex 会写代码就想重做软件，看到大模型会总结就想做知识库——方向都可以做，但接不上工作流和运营闭环，最后就停在 Demo。 06 ## 能快速 Demo，不意味着能进生产 AI 编程确实让开发变快了：以前一个小系统要排期，现在一个人几天就能做出像样的版本。但开发速度变快，不代表企业的工作方式自动变好。新系统上线后还有一连串问题：员工愿不愿意用？旧资料要不要搬？群聊里的确认、会议里的决策、审批任务日历还在原来的系统里怎么办？两年以后谁维护？ 很多企业最后得到一个尴尬结果：原来的系统还在，新做的软件也在，员工夹在中间来回搬运——这是用更快的方式，增加了一个新的工作地点。个人软件因为 AI 编程多了百倍，企业软件也多了好几倍，但真正被大量用户长期使用而稳定下来的寥寥无几。企业软件要的是稳定、可扩展、好维护，而不是只有你一家在用的 DIY 工具。 07 ## 企业需要的是办公环境里的 AI 企业每天的工作，本来就发生在办公协同软件里：人、组织、权限、文档、表格、任务、审批、日历、会议纪要和消息记录都在那里。它不只是聊天工具，更像企业日常运转的工作现场。AI 要发挥作用，就该进入这个现场——不是待在一个新系统里等员工打开，而是出现在员工本来就在工作的地方。 员工在群里发来一张发票，它能接住；顾问在表格里更新状态，它能看懂；纪要里出现待办，它能追踪；日报里出现异常，它能提醒负责人；结果出来，它写回原处。这也是我们把 Agent 框架放在办公协同软件之上的原因：办公软件提供工作环境，Agent 框架在里面部署能理解上下文、调用工具、推动任务的 Agent。传统软件让人进入系统；Agent 更应该进入人的工作流。 08 ## 一个简单的判断标准 未来几年，企业会看到很多快速开发、AI 编程、Agent 应用方案。一个方案如果只讲页面、功能、模块、数据库、上线，却说不清怎么进入企业每天的办公环境、怎么连接人、文档、表格、任务、审批、会议和已有系统，那它大概率只是一个更快的外包开发项目：能产生 Demo，不一定能改变工作方式。 问题一：它是不是在企业每天工作的地方运行？ 问题二：它能不能理解真实上下文？ 问题三：它能不能调用已有工具？ 问题四：它能不能推动任务继续往前走？ 问题五：它能不能把结果沉淀回原来的办公环境？如果回答不上来，它很可能只是一个新软件；能跑通，才开始像企业里的智能体。 ## 一手参考 原文 · 微信公众号「品哥聊AI」 ↗ FAQ · 快速回答 ## 能不能用 Claude Code 或 Codex 直接给企业重做一套 ERP / CRM？ 技术上能很快做出 Demo，但企业软件要的是稳定、可扩展、好维护。真实的卡点通常不在系统本身，而在系统之外的动作：资料在群聊里、确认在会议里、审批人临时换了。重做软件往往只是给企业多加一个需要维护、需要搬运数据的新地点。先问自己：要的是新软件，还是业务更流畅？ ## 怎么判断一个企业 AI 方案会不会停在 Demo？ 问五个问题：它是不是在企业每天工作的地方运行？能不能理解真实上下文？能不能调用已有工具？能不能推动任务往前走？能不能把结果沉淀回原来的办公环境？答不上来的，大概率只是一个更快的外包开发项目。 ## 企业 AI 落地应该分哪几层看？ 三层：工具层解决单点能力（识别发票、写代码、生成日报）；工作流层让 AI 在邮件、群聊、表格、审批之间连续做事；运营闭环层让异常被处理、过程被记录、经验进入复盘。只有工具层的方案，很难留在企业里。 PZ 周品 Pin Zhou 从业务判断、客户增长与组织流程出发，把 AI 连接到可验证的生产系统。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Enterprise AI Myth #1: Just Vibe-Code a New App with Codex or Claude | BEE Sigma URL: https://beesigma.com/insight-ai-myth-1-diy-software-en Summary: The old system is clunky and AI can write code now — so why not rebuild it? Enterprise AI Myth #1: Just Vibe-Code a New App with Codex or Claude | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Adoption Myths · Part 1 ## Enterprise AI Myth #1: Just Vibe-Code a New App with Codex or Claude Reviewed 21 Jul 20268 minPin Zhou 01 ## Do you want new software, or smoother business? More than one owner has asked me this year: our ERP and CRM are clunky — can you use AI to build us a replacement? The reasoning is understandable: the old system hurts, and Codex, Claude Code and Cursor can all write code now, so why not rebuild? Sounds smooth. I usually ask one question first: do you want a new piece of software, or do you want your business to run with less friction? Most owners pause. The two answers look similar and are worlds apart. Building software is just an upgrade of the old software, with custom features. Removing friction is about where the work happens every day, who handles it, where the material comes from, who takes over when it jams, and where the result lands — none of which necessarily requires new development. Most problems are not short of software. They are short of continuous action at the place of work: files live in group chats, confirmations happen in meetings, the spreadsheet is in another folder, the approver changed this week, the boss reshuffled priorities in one sentence, and someone forgot to update the status. A rebuilt system does not automatically absorb any of that — often it just adds one more node the company has to maintain. 02 ## Case 1 · Invoice approval at an engineering firm: one invoice drags a whole chain The traditional build would be an invoice management system: upload, extract fields, match the PO, run approval, land on a dashboard. The demo would look great. But the real problem starts the moment an invoice hits the inbox: the PO number matches no project — now what? One PO maps to several invoices — which are normal, which duplicated? Supplier names are written inconsistently — who confirms they are the same company? Amount and GST do not reconcile — return it, or send to finance for review? The project manager changed — who approves now? None of that lives on a page or in a recognition model. It is scattered across inboxes, project files, PO sheets, group-chat confirmations, finance reviews and approval records. Build a standalone invoice app and staff still carry material in, carry exceptions out, and ask around in the chat — the system decays into another spreadsheet. The effective move is to bring AI into the existing environment: as an invoice arrives it extracts fields, finds the project, links the PO, routes exceptions to the right person, and writes results back to the documents, sheets and approval records. Invoice approval is about how an invoice keeps moving forward after it enters the company. 03 ## Case 2 · Admissions at an education business: a CRM records state; deals do not advance themselves An admissions CRM sounds complete: lead pool, client status, funnel, dashboards. But admissions work does not happen only inside a CRM — conversations live in chat apps, documents in email, statuses in Excel, key clients get discussed in meetings, and plenty of judgment lives in counsellors’ heads. When a client stalls, the CRM says "following up" — and the owner reading those words still cannot tell whether the deal is in danger. Rebuild the CRM and the usual outcome is: counsellors fill in one more system, the boss watches one more dashboard, and real opportunities keep leaking. What helps is AI working where counsellors already work: reading statuses in Excel, checking chats for follow-ups, spotting missing documents in email, prompting the next action, surfacing at-risk clients. What admissions loses most often is the next step never being pushed in front of a person in time. 04 ## Case 3 · Store operations: a dashboard displays exceptions; it cannot get them handled Store projects turn into pretty systems easily: home page, charts, rankings, daily sales, stock alerts. Demos look great. But what an owner wants daily is not which number turned red — it is: who saw this exception? Who handles it? Was it handled? With what result? Will it happen again? A dashboard can only tell you a problem appeared; it cannot perform the actions that follow. Store operations close their loop better inside the daily work environment: the daily report generates, AI flags exceptions, exceptions route to owners, results write back into the report, and recurring issues get compiled for management. The worst outcome in store management: a beautiful dashboard, and nobody catching the actions. 05 ## The three cases, one structure: three layers Invoices, admissions, stores — three industries that decompose into the same three layers on the ground. The tool layersolves point capabilities — reading an invoice, generating a report, writing code. Coding agents, LLMs and OCR live here. It makes one action faster; it does not catch the whole chain. The workflow layeris where companies actually jam: after extraction, where is the PO checked? Who gets the exception? Where does the result write back? This happens between email, chats, sheets, approvals and meetings — the workflow layer makes AI act continuously across those nodes. The operating loopdecides whether AI stays: continuously spotting exceptions, driving owners to handle them, recording the process, feeding review — so the next run gets smoother. Handling one invoice once is worth little; entering operations is what counts. Why projects drift: they mistake tool-layer capability for the whole deployment. Codex writes code, so rebuild the software; the model summarises, so build a knowledge base — all workable directions that stall at Demo without the workflow and the loop behind them. 06 ## A fast demo is not production AI coding genuinely made development faster: what needed a project schedule can now be a passable version built by one person in days. But faster development does not automatically improve how a company works. After launch come the questions: will staff actually use it? Does old material migrate? What about confirmations in group chats, decisions in meetings, the approvals, tasks and calendars still living in the old systems? Who maintains it in two years? Many companies end up in an awkward spot: the old system still runs, the new software also runs, and staff shuttle data between them — a faster way to add one more place to work. Personal software has multiplied a hundredfold in the vibe-coding era, and business software severalfold; the number that stabilised under sustained real use is vanishingly small. Enterprise software needs stability, extensibility and maintainability — not a DIY tool used by exactly one company. 07 ## What companies need is AI inside the work environment Daily work already happens inside collaboration suites: people, org structure, permissions, documents, sheets, tasks, approvals, calendars, meeting minutes and message history all live there. That is not a chat tool — it is the shop floor of daily operations, and that is where AI should show up: not waiting inside a new system for staff to open it, but present where people already work. Someone drops an invoice into the group chat — it catches it. A counsellor updates a sheet — it understands. A to-do appears in the minutes — it tracks it. An anomaly shows in the daily report — it alerts the owner. When results land, it writes them back. That is why we put the agent framework on top of the collaboration suite: the suite provides the environment; the framework deploys agents that understand context, call tools and push tasks. Traditional software makes people enter the system. Agents should enter people’s workflow. 08 ## A simple test You will see plenty of rapid-build, AI-coding, agent-application pitches in the coming years. If a proposal talks pages, features, modules, databases and launch — but cannot explain how it enters your daily work environment and connects people, documents, sheets, tasks, approvals, meetings and existing systems — it is most likely just a faster outsourced dev project. It can produce a demo. It will not necessarily change how work gets done. Question 1: does it run where the company already works every day? Question 2: can it understand real context? Question 3: can it call the tools you already have? Question 4: can it push tasks forward? Question 5: can it write results back into the original environment? If these cannot be answered, it is probably just new software. If they run, it starts to be an agent inside the business. ## Primary sources Original article · WeChat 「品哥聊AI」 ↗ FAQ · Quick answers ## Can Claude Code or Codex just rebuild our ERP / CRM? Technically a demo comes fast — but enterprise software needs stability, extensibility and maintainability. The real blockers usually sit outside the system: files in group chats, confirmations in meetings, approvers changing mid-week. A rebuild often just adds one more place that needs maintenance and manual data shuttling. Ask first: do you want new software, or smoother business? ## How do I tell whether an enterprise AI proposal will stall at the demo? Five questions: does it run where you already work every day? Does it understand real context? Can it call your existing tools? Can it push tasks forward? Can it write results back into the original environment? If those cannot be answered, it is likely a faster outsourced dev project. ## What are the layers of an enterprise AI deployment? Three: the tool layer for point capabilities (reading invoices, writing code, generating reports); the workflow layer where AI acts continuously across email, chats, sheets and approvals; and the operating loop where exceptions get handled, processes recorded and lessons reviewed. Tool-layer-only solutions rarely stay. PZ Pin Zhou Connecting AI to verifiable production systems — starting from business judgment, client growth and organizational process. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 企业 AI 落地误区之二：以为建一个本地知识库，就能沉淀企业知识 | BEE Sigma URL: https://beesigma.com/insight-ai-myth-2-knowledge-base Summary: 文件进了服务器，知识还没有进入工作。知识库有三大坎：没人建、没人用、没人维护。真实数据：377 条发票试运行 95.49% 正确率，卡住的七成不是 AI 读不懂，而是规则和数据不清。企业缺的不是知识库，是知识流。 企业 AI 落地误区之二：以为建一个本地知识库，就能沉淀企业知识 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 落地误区 · 系列二 ## 企业 AI 落地误区之二：以为建一个本地知识库，就能沉淀企业知识 内容复核 2026.07.2712 分钟周品 Pin Zhou 01 ## 先看看知识库的鼻祖 中国古代最著名的“知识库管理者”，可能是孔子。他自称“述而不作，信而好古”，整理并传授《诗》《书》《礼》《乐》《易》《春秋》——替一个文明整理出了底层知识库。但真正体现他如何深度运用知识的，是《论语》：面对具体的人和问题，现场组织知识、生成答案。 硬套今天的说法：六经是知识底座，弟子的提问是 Prompt，孔子负责理解提问者、判断处境、组织上下文，再从经典与经验中调用知识，生成针对当下的回答——《论语》就是这套系统一次次真实运行后留下的结果。 知识库不会自己变聪明。只有进入具体场景，流动的知识才有价值。流水不腐。 02 ## 文件进了服务器，知识还没有进入工作 遇到有“知识库”想法的老板，我通常先追问三句：你们现在每次会议都有纪要吗？项目状态有人持续更新吗？合同加了补充协议，能自动回到同一个客户和项目下面吗？如果答案都是否，那台本地服务器很快就会无人问津——还记得前两年全中国都在卖的一体机吗？ 两年前做 2Brain 知识库的时候我就发现：AI 确实极大提高了知识生产效率，一个原来不爱写纪要的人，现在十分钟能生成三千字。过去是没人写，现在是 AI 写得太多，没人看。 三大坎：没人建、没人用、没人维护。这三个坎不会因为 AI 生成快而消失，反而因为内容太多，大幅增加了验证、迭代、对齐的维护成本。 真正的误区和本地部署还是云端部署关系不大——企业每天最有价值的信息，是 ERP 里刚变化的库存、CRM 里刚流失的客户、一段客服电话、一张现场照片、一次设备报警。 它们还算不上知识：只有当库存变化对应到订单、投诉对应到产品、报警找到责任人和历史处理记录，这些散落的事实才形成可判断、可行动的业务上下文。 03 ## 大部分知识库，只装进了已经写成文件的那部分 企业真正运转依赖的，是那些不断发生、不断变化、很难及时写进文档的事实，以及事实之间的关系。Palantir 把这一层叫 Ontology（本体）：客户、订单、合同、设备、员工被定义成业务对象，对象之间有关系，关系背后有规则，规则触发动作——一个持续反映企业真实运行状态的业务世界。 过去做这件事非常贵。OpenClaw 这类 Agent 框架出现后，AI 调用不同系统、读取多种模态、按任务动态组织上下文的门槛大幅下降。但它不会凭空理解一家企业：客户是谁、订单属于哪个项目、谁有权限、异常触发什么动作，这些业务语义仍要有人定义。Ontology 解决“企业里的东西是什么、有什么关系”；Agent 框架解决“眼前这个任务该调哪些信息和工具”。两层打通，知识才流动。 04 ## 一个建筑客户的试运行数字 A 公司的发票全从邮件进来，财务逐张认供应商、发票号、金额、GST、项目、PO、审批人。客户一开始想要两样东西：把 PDF 存进知识库方便查，用 OCR 减少录入。我们没先建库，而是把发票邮件、PDF 识别、项目规则、审批节点和付款状态接成一条工作流。 试运行数字：一轮 377 条真实记录，初步处理正确率 95.49%；另一轮 407 张候选发票里，OCR 真正识别失败的只有 12 张（2.9%）。 卡住的大头在别处：规则不完善 173 张，供应商映射、PO、必要字段缺失 132 张——AI 读不懂的不到 3%，规则和数据不清楚的占了七成多。 客户第一次看清的其实是自己：哪些规则从来没人写下来，哪些基础数据缺胳膊少腿，哪条责任链没有闭环。这些东西放十台本地服务器也存不出来，流程真跑起来才会现形。 05 ## 过去，知识为什么流不起来 过去的企业软件有个默认前提：一个系统处理一类数据，一个场景解决一个问题。但一件真实业务很少只活在一种系统里。客户投诉一批设备：合同里有验收标准，ERP 里有生产批次，现场有故障照片，客服录音里有描述，车间视频里有操作画面。过去把这些放进同一次判断，成本高到没人愿意折腾。 这一轮多模态 AI 最大的变化，是孤岛之间终于开始修桥：AI 能在同一个任务里读合同、看表格、理解图片、听录音。但能看懂多种模态只相当于有了五官——能不能认出它们讲的是同一个客户、同一笔订单、同一次故障，才决定 AI 有没有真正的业务理解。 06 ## 三个根本问题 第一，它知不知道今天发生了什么。AI 最危险的时候不是胡说，而是用完整的结构、专业的语气，引用一份已经失效的制度，给你一个过期的答案。企业里每个回答都该带着时间：基于哪个版本？数据更新到什么时候？ 第二，它能不能认出这些信息讲的是一件完整的事。上传合同问付款条件，答得很好；但项目经理真正的问题是：这笔款今天该不该催？验收了没有，发票开了没有，ERP 里到账了没有？只有把不同系统、不同模态的信息对齐到客户、项目、订单、设备和责任人，知识才有骨架。 第三，它能不能把事情继续往前推。开完会，纪要没有变成任务；合同执行了，风险没人提醒。这个 AI 很勤奋，但只负责发言，不负责交付。我们给新西兰一家房屋贷款公司（B 公司）接的流程，把线索进来的第一分钟用了起来：客户从微信进来——首套买房、预算 89 万纽币、这周拍卖、deposit 只有 15%——系统认出来源与紧急度、标出风险、对照档案排重，按内部标准直接分给合适的顾问，附上下一步动作建议。顾问打开的是一张能直接行动的客户卡。回答之后有判断，判断之后有任务，任务落到人头上，这才叫往前推：信息 → 判断 → 任务 → 行动 → 结果 → 反馈。 07 ## 企业知识，是在结果回来以后形成的 再说一个零售电商客户（C 公司）。财务对账原来全靠 Excel 和人肉。我们第一期边界划得很死：不改原业务系统、不自动发客户、不碰审批流、不重做 ERP，只加了两个指令——一个生成我方标准对账单，一个对客户回传表做匹配与差异校验、输出异常清单。 关键动作在最后一步：每一轮对账处理完的异常，反过来沉淀成规则——字段怎么映射、金额差异怎么归因、退单怎么抵消、哪类问题找销售确认。第一期就沉淀出十种异常类型；异常类型稳定后，下一轮直接复用规则，省掉重新人工判断。被业务结果反复验证过的判断，才更接近企业真正的资产。 08 ## 本地化解决归属与安全，知识流形成闭环 写到这里若被理解成反对本地知识库，就误会了。合同、研发资料、图纸、客户档案，这些数据本来就该掌握在企业自己手里，有些甚至不该离开内网。本地部署解决的是：数据在哪里、归谁控制、哪些内容可以离开内网、访问能不能被审计——这些是基础设施，该建，而且要预留和内部系统打通的接口。 知识能不能流动是另一层问题：分散在本地服务器、SaaS、业务数据库和边缘设备里的信息，能不能在任务发生时一起工作。对中小企业，不需要模仿 Palantir 的 AIP 和 Ontology 抽象层——这柄剑你可能舞不动。正确做法是让数据、信息、知识留在原来的系统里，不追求覆盖取代；在我们数十个成功落地的方案里，没有给任何客户开发新的 SaaS 逼员工使用。真正该做的，是把内容对齐到正确的客户、项目、订单和事件，建立连接、身份、权限、关系、记忆和长期上下文，并把这套规则做到可以交给你维护。 很多老板以为企业知识像一个仓库，资料放得越多，企业越聪明。我更愿意把它看成一条河：仓库只关心东西有没有入库，河流还要关心水从哪里来、流向哪里、中间有没有堵住、最后能不能回到系统里。那些经过一次次流动、最终留下来的东西，才配叫沉淀。 ## 一手参考 原文 · 微信公众号「品哥聊AI」 ↗ FAQ · 快速回答 ## 企业建知识库最常见的坑是什么？ 三大坎：没人建、没人用、没人维护。AI 让知识生产变快之后反而更难——过去是没人写，现在是 AI 写得太多没人看，验证、迭代、对齐的维护成本大幅上升。 ## 本地部署知识库还有必要吗？ 有必要，但它解决的是数据在哪里、归谁控制、能否审计这类归属与安全问题。知识能不能流动是另一层：分散在本地服务器、SaaS、业务数据库里的信息，能不能在任务发生时一起工作。两层都要，别混为一谈。 ## 知识流和知识库有什么区别？ 知识库关心资料有没有入库；知识流关心“信息 → 判断 → 任务 → 行动 → 结果 → 反馈”这条链有没有跑通。被业务结果反复验证过的判断，才是企业真正的资产。 ## 卡住企业知识库的，真是 AI 能力不够吗？ 通常不是。一个真实试运行：407 张发票里 OCR 真正识别失败的只有 12 张（2.9%）；规则不完善的 173 张、供应商映射与必要字段缺失的 132 张——七成以上卡在规则和基础数据，流程真跑起来这些问题才会现形。 PZ 周品 Pin Zhou 从业务判断、客户增长与组织流程出发，把 AI 连接到可验证的生产系统。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Enterprise AI Myth #2: A Local Knowledge Base Means Your Knowledge Is Captured | BEE Sigma URL: https://beesigma.com/insight-ai-myth-2-knowledge-base-en Summary: Files on a server are not knowledge at work. Knowledge bases face three walls: nobody builds, nobody uses, nobody maintains. Enterprise AI Myth #2: A Local Knowledge Base Means Your Knowledge Is Captured | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Blog · Adoption Myths · Part 2 ## Enterprise AI Myth #2: A Local Knowledge Base Means Your Knowledge Is Captured Reviewed 27 Jul 202612 minPin Zhou 01 ## Start with the original knowledge manager The most famous "knowledge base manager" in Chinese history may be Confucius. He compiled and taught the Six Classics — the knowledge foundation of a civilisation. But what shows how he actually used knowledge is the Analects: facing a specific person with a specific problem, organising knowledge on the spot and generating an answer. In today’s terms: the Six Classics were his knowledge base, a student’s question was the prompt, and Confucius understood the asker, judged the situation, assembled context and generated an answer for that moment. The Analects is the log of that system running, again and again, on real problems. A knowledge base does not get smarter by itself. Knowledge only earns value when it flows into real situations. Still water goes stale. 02 ## Files reached the server; knowledge never reached the work When an owner brings up "knowledge base", I ask three questions first: does every meeting produce minutes? Does anyone keep project status current? When a contract gains an addendum, does it automatically attach to the same client and project? If the answers are no, that local server will soon gather dust — remember the wave of all-in-one LLM appliances a couple of years back? Building the 2Brain knowledge product two years ago taught me something uncomfortable: AI massively increased knowledge production. Someone who never wrote minutes can now generate three thousand words in ten minutes. It used to be that nobody wrote. Now AI writes too much, and nobody reads. Three walls: nobody builds, nobody uses, nobody maintains. Faster generation does not remove them — it raises the cost of verifying, iterating and aligning all that content. The real misconceptionhas little to do with on-premise versus cloud. A company’s most valuable information today is the stock level that just changed in the ERP, the client who just went cold in the CRM, a service call recording, a site photo, an equipment alarm. Those are not yet knowledge: only when the stock change maps to an order, the complaint maps to a product, and the alarm finds its owner and past handling records do scattered facts become business context you can judge and act on. 03 ## Most knowledge bases only hold what was already written down What actually runs a business is the stream of facts that keep happening and changing — and the relationships between them. Palantir calls this layer the Ontology: clients, orders, contracts, equipment and staff defined as business objects; objects carry relationships; relationships carry rules; rules trigger actions — a living model of how the business actually runs. That used to be extremely expensive. Agent frameworks like OpenClaw have collapsed the cost of calling different systems, reading multiple modalities and assembling context per task. But no framework understands your company out of thin air: who the client is, which project an order belongs to, who may see what, what an exception should trigger — someone still has to define the semantics. The ontology answers "what things are and how they relate"; the agent framework answers "for this task, what to pull and what to do next". Connect the two layers and knowledge starts to flow. 04 ## The pilot numbers from a construction client Company A’s invoices all arrive by email as PDFs. Finance identified supplier, invoice number, amount, GST, project, PO and approver by hand. The client initially wanted two things: store the PDFs in a searchable knowledge base, and use OCR to cut typing. We did not build the library first — we wired invoice email, PDF recognition, project rules, approval nodes and payment status into one workflow. Pilot numbers: one round of 377 live records ran at 95.49% first-pass accuracy; in another round of 407 candidate invoices, true OCR failures numbered just 12 (2.9%). The real blockers were elsewhere: 173 invoices hit incomplete rules; 132 hit missing supplier mappings, POs or required fields. AI misreads: under 3%. Unclear rules and data: over 70%. What the client finally saw was itself: which rules had never been written down, which base data was missing limbs, which responsibility chain never closed. Ten local servers cannot store that — it only surfaces when the process actually runs. 05 ## Why knowledge never flowed before Legacy enterprise software assumed one system per data type, one scenario per problem. But a real piece of business rarely lives in one system. A client complains about a faulty batch: the contract holds acceptance criteria, the ERP holds batch and shipping records, the site team has photos, the service recording holds the client’s account, the workshop video may hold the operation itself. Bringing all of that into a single judgment used to cost more than anyone would pay. The biggest change from multimodal AI is that bridges are finally being built between the islands: one task can now read the contract, scan the table, understand the photo and listen to the call. But reading many modalities only gives AI senses. Recognising that they all describe the same client, the same order, the same failure — that is what real business understanding means. 06 ## Three fundamental questions First: does it know what happened today? AI is most dangerous not when it hallucinates, but when it cites a policy that expired — in perfect structure and a professional tone. Every answer inside a company should carry time: based on which version? Data current to when? Second: can it recognise one complete event? Upload a contract and it answers payment terms beautifully. But the project manager’s real question is: should this payment be chased today? Was acceptance signed, was the invoice issued, did the money land in the ERP? Only when information from different systems and modalities is aligned to clients, projects, orders, equipment and owners does knowledge grow a skeleton. Third: can it push things forward? For a New Zealand mortgage brokerage (Company B), we made the first minute of every lead count: an enquiry arrives on WeChat — first home, NZD 890k budget, auction this week, only 15% deposit. The system identifies source and urgency, flags the risks, deduplicates against records, routes to the right adviser by the firm’s own playbook, and attaches the next actions. The adviser opens a client card they can act on immediately. Answer leads to judgment, judgment to task, task to a named owner — that is pushing forward: information → judgment → task → action → result → feedback. 07 ## Company knowledge forms after the results come back A retail e-commerce client (Company C) reconciled accounts with Excel and manual labour. We drew hard boundaries for phase one: no changes to core systems, nothing auto-sent to clients, no touching approvals, no ERP rebuild — just two commands: one generates our standard statement, one matches the client’s returned sheet and outputs a discrepancy list. The decisive move was the last step: every round’s handled exceptions crystallise into rules — how fields map, how amount differences attribute, how returns offset, which issues go to sales for confirmation. Phase one alone produced ten stable exception types; once stable, the next round reuses the rules and skips fresh manual judgment. Judgments repeatedly verified by business results are what company assets actually look like. 08 ## On-premise solves ownership and security; the flow closes the loop None of this argues against local deployment. Contracts, R&D files, drawings, client records — that data should be in the company’s own hands, some of it never leaving the intranet. On-premise answers: where data lives, who controls it, what may leave, whether access can be audited. Build that — with interfaces reserved for your internal systems. Whether knowledge flows is a different layer: can information scattered across local servers, SaaS tools, business databases and edge devices work together when a task happens? For SMEs, do not imitate Palantir’s AIP and ontology abstractions — that sword may be too heavy to swing. Keep data where it already lives, replace nothing; across dozens of client deployments we have never built a new SaaS and forced staff into it. The real work is aligning content to the right client, project, order and event — building connection, identity, permissions, relationships, memory and long-term context — then handing you a rule set you can maintain. Many owners picture company knowledge as a warehouse: the more you store, the smarter you get. I prefer a river: a warehouse only cares whether things checked in; a river cares where the water comes from, where it flows, whether it jams, and whether it returns to the system. What survives round after round of flowing — that deserves to be called an asset. ## Primary sources Original article · WeChat 「品哥聊AI」 ↗ FAQ · Quick answers ## What are the most common failure modes when building an enterprise knowledge base? Three walls: nobody builds it, nobody uses it, nobody maintains it. Faster AI generation makes it harder, not easier — it used to be that nobody wrote; now AI writes too much and nobody reads, while the cost of verifying and aligning content climbs. ## Is an on-premise knowledge base still worth it? Yes — for ownership and security: where data lives, who controls it, whether access is auditable. Whether knowledge flows is a separate layer: can information scattered across servers, SaaS and databases work together when a task happens. You need both; do not conflate them. ## What is a knowledge flow versus a knowledge base? A knowledge base cares whether material was filed. A knowledge flow cares whether the chain runs: information, judgment, task, action, result, feedback. Judgments repeatedly verified by business results are the real company asset. ## Is AI capability really what blocks enterprise knowledge bases? Usually not. In one live pilot, only 12 of 407 invoices (2.9%) were true OCR failures — while 173 hit incomplete rules and 132 hit missing mappings or fields. Over 70% of blockers were rules and base data, and they only surface when the process actually runs. PZ Pin Zhou Connecting AI to verifiable production systems — starting from business judgment, client growth and organizational process. Read Next Methodology · Boss AI cloneReviewed 14 Oct 20269 min ## What Is a Boss AI Clone? How to Turn a Business Owner’s Judgement into an AI Agent Your Team Can Ask What is a boss AI clone, how is it different from ChatGPT, how is one built, how long does it take, how is it installed in Lark or WeCom, and what can it not do? Answered with the BossClone six-layer process, walkthrough screenshots and records from six clients. Read article Methodology · Efficiency & ProcessReviewed 6 Oct 20268 min ## How Businesses Use AI to Improve Efficiency: Five Workflow Levers From Dated Project Records How do businesses use AI to improve efficiency? Not by replacing software, but by connecting AI agents to the repetitive, rule-based, data-ready steps inside existing processes. Using four BEE Sigma projects from 2026 with dated records (91.48% invoice routing, site timesheets in 10 weeks, sales reconciliation in 11 days, cross-border orders in 90 days), this article sets out five efficiency levers: document routing, field data entry, cross-system reconciliation, customer orders and market reach, with what the AI does, what people do and how long each takes. Read article ## Hand your first workflow to the agents One free scan shows how visible you are in the AI era; the AIM assessment finds your best angle of adoption. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # 为什么企业 AI 项目停在 Demo：缺的通常不是模型 | BEE Sigma URL: https://beesigma.com/insight-demo-to-production Summary: AI 从演示走到生产落地，要同时满足业务责任、可用输入、人工边界、系统连接与可验收指标。这是 BEE Sigma 落地团队整理的五项上线判断标准。 为什么企业 AI 项目停在 Demo：缺的通常不是模型 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 实施方法 ## 为什么企业 AI 项目停在 Demo：缺的通常不是模型 内容复核 2026.08.136 分钟BEE Sigma 落地团队 01 ## Demo 回答“能不能”，生产系统回答"谁负责" Demo 可以从一份干净资料和一个理想问题开始。真实业务会同时出现旧版本、权限差异、缺失字段、例外情况和时间压力。模型输出只是流程中的一步。 因此第一个设计问题不是选哪个模型，而是明确流程所有者、可接受错误、必须人工复核的节点，以及失败时如何回退。 02 ## 用一个最小业务闭环验证，而不是做功能清单 一个有效的验证应从真实输入开始，进入真实岗位任务，产生可以被业务人员判断的输出，并把结果回写到下一步。只展示聊天界面，无法证明流程可以持续运行。 写清基线：现在花多久、由谁做、什么算失败。 写清边界：AI 可做、需复核、不得做。 写清门槛：达到什么条件继续投入，什么条件停止。 03 ## 上线只是证据收集的开始 生产系统需要持续观察采用率、人工覆盖、异常类型、答案质量和业务结果。技术可用但无人采用，不是成功；采用率高但高风险错误不可控，也不是成功。 FAQ · 快速回答 ## AI 演示很惊艳，为什么一到生产环境就不行了？ Demo 回答的是“能不能”，生产系统回答的是“谁负责”：数据从哪来、异常谁接手、错误怎么回退、结果写回哪里。演示环境里没有这些约束，生产环境里全是。验收时别看演示效果，看闭环有没有跑通。 ## 企业上 AI 应该先做功能清单吗？ 不建议。先选一个最小业务闭环（例如一条跟单流或一条对账流），在真实数据、真实人员、真实异常里跑通，再逐步扩展。功能清单容易堆出一个什么都有、什么都没人用的系统。 ## AI 项目上线后就算完成了吗？ 上线只是证据收集的开始：命中率、异常率、人工接管率、处理时长，这些数据决定下一步是扩展还是回退。没有运行证据的“已上线”，和 Demo 没有本质区别。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # 企业知识库不能只看“能回答”：六个上线检查点 | BEE Sigma URL: https://beesigma.com/insight-knowledge-base Summary: 企业知识库能否在真实业务中被信任，取决于来源、版本、权限、引用、拒答与维护责任六件事。这是 BEE Sigma 编辑部整理的知识系统落地检查清单。 企业知识库不能只看“能回答”：六个上线检查点 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 知识资产 ## 企业知识库不能只看“能回答”：六个上线检查点 内容复核 2026.08.137 分钟BEE Sigma 编辑部 01 ## 检索准确不等于业务正确 系统可能找到相关段落，却忽略文件版本、适用地区、客户权限或例外条款。评测必须覆盖真实任务，不只问资料里最容易回答的问题。 02 ## 六个上线检查点 下面六项应在验证阶段形成可复用测试集，并在知识更新后重复运行。 来源覆盖：关键资料是否进入，缺失是否可见。 版本判断：系统能否识别当前有效内容。 权限：用户只能检索和生成其有权查看的信息。 引用：关键结论可以回到具体来源位置。 拒答与转交：证据不足时不编造，并知道交给谁。 维护责任：每类知识都有更新人和复核周期。 03 ## 把不确定性做成界面，而不是藏起来 可信系统会区分确定答案、需要补资料、需要专家复核和不得回答。对企业用户而言，这种边界比一个永远自信的回答框更有价值。 FAQ · 快速回答 ## 企业知识库检索很准，为什么员工还是不信任它？ 检索准确不等于业务正确：答案可能来自一份已过期的制度、一个不适用的版本。回答要带来源、版本与时间，拿不准时明确说“不确定”并转人工——把不确定性做成界面，而不是藏起来。 ## 企业知识库上线前应该检查什么？ 六个检查点：来源可溯、版本可控、权限对齐、拒答路径、人工兜底、更新机制。缺任何一个，知识库越权威，错误传播越快。 ## 知识库答错了谁负责？ 设计阶段就要定：高影响回答必须给出处，关键决策保留人工确认，错误反馈有入口、有人处理。没有责任设计的知识库，出错时只能怪“AI 不靠谱”。 BS BEE Sigma 编辑部 按服务品牌的实践内容呈现，来自 BEE Sigma 的实施方法与证据原则。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # 把智能体接入 CRM、ERP 与协作系统之前，先画清这张图 | BEE Sigma URL: https://beesigma.com/insight-connect-systems Summary: 有效的 AI 系统集成不是 API 数量竞赛，而是把数据、权限、动作、人工确认、日志与回退放在同一张流程图上。这是 BEE Sigma 落地团队的系统连接方法。 把智能体接入 CRM、ERP 与协作系统之前，先画清这张图 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 系统集成 ## 把智能体接入 CRM、ERP 与协作系统之前，先画清这张图 内容复核 2026.08.136 分钟BEE Sigma 落地团队 01 ## 先从业务动作出发 "接入 CRM"不是一个完整需求。真正的问题是：谁在什么条件下读取哪些字段，生成什么建议，谁确认后写回哪个对象，失败时留下什么状态。 02 ## 六层连接图 对每一个自动或半自动动作，至少画清以下六层。 触发：什么事件启动流程。 上下文：哪些数据可读，来源和版本是什么。 判断：规则与模型各负责什么。 确认：谁必须在执行前复核。 动作：写入、发送或调用的最小权限是什么。 证据：记录什么，如何回退和复盘。 03 ## 连接器数量不是生产能力证明 一个接口端点、一个演示连接和一个经过权限、异常、容量与运维验证的投产集成不是同一件事。对外说明时，应分别列出可用端点、可配置连接器与已投产集成。 FAQ · 快速回答 ## 把 AI 接入企业现有系统，需要更换 ERP 或 CRM 吗？ 不需要。正确路径是从一个业务动作出发（如“发票进来后自动关联 PO”），沿身份、权限、数据、动作、回写、审计六层把这条链接通——系统留在原位，AI 进入流程。 ## 供应商宣传支持几百个连接器，可信吗？ 连接器数量不是生产能力的证明。要问的是：这条连接在你的权限体系下能不能读写？异常时怎么回退？操作留不留审计痕迹？一条能落地的连接，胜过一百个演示用的连接器。 ## 接入现有系统最容易踩的坑是什么？ 跳过身份与权限层直接读写数据。短期见效快，长期造成越权访问与审计缺口。先定义“谁能以什么身份做什么”，再谈自动化。 DT BEE Sigma 落地团队 把工作流真正接进企业现有系统的一线交付团队，方法来自已交付项目。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # 新西兰企业启动 AI 工作流前，应先回答的七个治理问题 | BEE Sigma URL: https://beesigma.com/insight-responsible-ai-nz Summary: 新西兰企业启动 AI 工作流前应先回答七个治理问题：目的、个人信息、供应商、准确性、人审、透明度与持续复盘，把治理放进实施流程，而不是上线后的附录。 新西兰企业启动 AI 工作流前，应先回答的七个治理问题 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 博客 · 负责任 AI ## 新西兰企业启动 AI 工作流前，应先回答的七个治理问题 内容复核 2026.08.138 分钟BEE Sigma 编辑部 01 ## 治理从目的和数据最小化开始 先写清流程为什么需要 AI、不用 AI 是否也能解决、输入中有哪些个人或敏感信息，以及每一项数据是否真的必要。范围越模糊，后续供应商、权限与留存决策越难。 02 ## 七个上线前问题 这些问题不是法律意见，而是一个实施团队应当能用证据回答的运行清单。 目的：这个流程解决什么问题，成功与停止标准是什么？ 信息：哪些个人信息进入，能否减少或去标识？ 供应商：数据去哪里、保留多久、是否用于训练？ 准确性：用什么真实测试集评估，谁批准上线？ 人审：哪些决定必须由具备责任的人确认？ 透明度：客户和员工何时需要知道 AI 参与？ 持续复盘：如何处理投诉、异常、知识过期和供应商变化？ 03 ## 把 PIA 当作活的设计文件 新西兰隐私专员建议在使用 AI 处理个人信息前进行并持续更新隐私影响评估。实践上，这意味着流程、字段、供应商或用途变化时，评估也要更新。 ## 一手参考 新西兰隐私专员：AI 指南 ↗MBIE：AI 战略与企业指南 ↗ FAQ · 快速回答 ## 新西兰企业用 AI 需要做 PIA 吗？ 隐私专员建议在用 AI 处理个人信息前进行隐私影响评估（PIA），并把它当作活的设计文件——流程、字段、供应商或用途变化时同步更新，而不是一次性交差。 ## 负责任 AI 会不会拖慢落地速度？ 恰恰相反。目的限定、数据最小化、人审确认点这些设计，会在上线前逼出关键问题的答案；带着答案上线的系统返工更少、验收更快，也更敢按结果负责。 ## AI 治理应该写在哪份文件里？ 写进流程而不是政策文件里：权限在系统里生效、确认点在工作流里出现、日志在运行中留存。政策文件里的治理没人执行，流程里的治理不需要提醒。 BS BEE Sigma 编辑部 按服务品牌的实践内容呈现，来自 BEE Sigma 的实施方法与证据原则。 Read Next · 相关阅读 方法论 · 老板分身内容复核 2026.10.149 分钟 ## AI 老板分身是什么？怎么做：把老板的判断变成团队随时能问的智能体 AI 老板分身是什么、和 ChatGPT 有什么区别、怎么做、要多久、怎么装进飞书或企业微信、能做什么不能做什么？本文用 BossClone 的六层训练流程、实操截图和六家客户的使用记录，一次讲清楚。 阅读全文 方法论 · 效率与流程内容复核 2026.10.068 分钟 ## 企业如何利用 AI 提升效率：优化工作流程的五个有记录的杠杆 企业如何利用 AI 提升效率？答案不在换一套软件，而在把 AI Agent 接进现有流程里重复、规则清楚、数据现成的那几步。本文用 BEE Sigma 2026 年四个有日期记录的项目（发票分流 91.48%、工地工时 10 周、销售对账 11 天、跨境订单 90 天）拆出五个效率杠杆：单据分流、现场录入、跨系统对账、客户订单、市场触达，每个说明 AI 做哪步、人做哪步、多久见效。 阅读全文 ## 把第一条业务流，交给 Agent 试试 一次免费诊断，看清企业在 AI 时代的可见度；AIM 评估帮你找到最合适的落地切入角度。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # 负责任 AI 实施 | BEE Sigma URL: https://beesigma.com/responsible-ai Summary: 治理要进入流程，而不是留在政策文件里。BEE Sigma 把目的、数据流、权限、来源、人审、日志、异常与供应商边界纳入设计。 负责任 AI 实施 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 Responsible AI · 负责任 AI ## 治理要进入流程，而不是留在政策文件里 BEE Sigma 的实施方法建议把目的、数据流、权限、来源、人审、日志、异常与供应商边界纳入设计。落到执行上是 6 项治理原则、1 张 4 步关键数据流图、3 种部署形态（客户云 / 本地离线 / 托管），具体控制取决于项目风险、技术与合同范围。 01 ## 目的与最小必要 我们的项目方法建议先限定业务目的，再评估哪些数据可以减少、去标识或不收集。 02 ## 来源与准确性 关键回答应尽量保留来源、版本与测试证据；证据不足时设计拒答或转交路径。 03 ## 人审与责任 对高影响决定，我们建议明确具备相应职责的确认人，并在设计中保留责任边界。 04 ## 权限与隔离 项目通常按角色、客户与任务评估隔离要求，并以最小权限作为设计目标。 05 ## 日志、异常与回退 根据风险与合同范围，保留必要运行证据，并设计异常、人工处理与回退路径。 06 ## 持续复盘 当模型、知识、供应商或用途变化时，应重新评估；一次验收不能代表永久保证。 Data Flow · 绘制关键数据流 ## 让重要步骤可解释、可复核 我们建议在项目范围内记录关键输入、处理、复核与保留步骤。具体供应商、区域、保留期与训练用途必须按项目和合同确认。 01输入来源、目的、必要性 02处理模型、规则、供应商 03复核责任人、异常、拒答 04保存 / 删除期限、日志、退出 Deployment · 部署不是一句"私有化" ## 私有化还是托管？按数据、控制与运维责任怎么选？ A ## 客户云 / 受控云 在客户认可的云与区域中运行，明确身份、日志、密钥与供应商责任。 B ## 本地或离线 只有在模型、硬件、更新、运维与功能限制明确时，才使用“离线”表述。 C ## 托管服务 明确数据流、子处理方、保留、退出与服务责任；便捷不替代审查。 NZ · NEW ZEALAND CONTEXT ## 新西兰企业用 AI，为什么需要持续更新的隐私影响评估？ 新西兰《隐私法 2020》（Privacy Act 2020）自 2020 年 12 月 1 日生效，含 13 条信息隐私原则；隐私专员建议在使用 AI 处理个人信息前进行并持续更新隐私影响评估（PIA）。流程、字段、供应商或用途变化时，评估也应随之更新。 ## 把第一条业务流，交给 Agent 试试 先判断哪里值得做，再决定用什么 AI。一次免费诊断 + 一次落地评估，把治理放进第一天的设计。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Responsible AI Implementation | BEE Sigma URL: https://beesigma.com/responsible-ai-en Summary: Governance belongs inside the process, not in a policy binder. Responsible AI Implementation | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan Responsible AI ## Governance belongs in the process, not in a policy binder Responsible AI here means governance built into the workflow, not a policy binder. BEE Sigma’s implementation method designs in purpose, data flow, permissions, sources, human review, logging, exceptions and vendor boundaries. In practice that means 6 governance principles, 1 four-step critical data-flow map, and 3 deployment forms (customer cloud / on-premise offline / managed). Specific controls depend on project risk, technology and contract scope. 01 ## Purpose & data minimization Scope the business purpose first, then assess what data can be reduced, de-identified or simply not collected. 02 ## Sources & accuracy Key answers keep their sources, versions and test evidence wherever possible; when evidence is thin, design refusal or hand-off paths. 03 ## Human review & accountability For high-impact decisions, name an accountable approver and keep the responsibility boundary in the design. 04 ## Permissions & isolation Assess isolation by role, client and task — with least privilege as the design target. 05 ## Logs, exceptions & rollback Keep the operating evidence the risk and contract require; design exception handling, human fallback and rollback paths. 06 ## Ongoing review Re-assess when models, knowledge, vendors or purpose change; one acceptance test is never a permanent guarantee. Data Flow · Map the critical path ## Make the important steps explainable and reviewable Within project scope, record the key input, processing, review and retention steps. Specific vendors, regions, retention periods and training usage must be confirmed per project and contract. 01Inputsource, purpose, necessity 02Processingmodels, rules, vendors 03Reviewowner, exceptions, refusals 04Retain / Deleteduration, logs, exit Deployment · More than the word “private” ## Private or managed? How do you choose by data, control and operational responsibility? A ## Customer / controlled cloud Run in a cloud and region the client approves, with identity, logging, keys and vendor responsibility made explicit. B ## On-premise or offline Use the word “offline” only when models, hardware, updates, operations and feature limits are all explicit. C ## Managed service Make data flow, sub-processors, retention, exit and service responsibility explicit; convenience never replaces scrutiny. NZ · NEW ZEALAND CONTEXT ## Why do New Zealand businesses need a continuously updated Privacy Impact Assessment? A Privacy Impact Assessment (PIA) is a structured review of how a system collects, uses and retains personal information. New Zealand’s Privacy Act 2020 took effect on 1 December 2020 and sets out 13 information privacy principles. The Privacy Commissioner recommends conducting — and continuously updating — a PIA before using AI on personal information. When the process, fields, vendors or purpose change, so should the assessment. ## Hand your first workflow to the agents Decide where it’s worth doing before deciding which AI. One free scan plus one adoption assessment — governance designed in from day one. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI ===== # Local Growth Launch · 本地获客启动包 | BEE Sigma URL: https://beesigma.com/local-growth-launch Summary: 面向本地服务型中小企业（以英语市场为主，全球远程交付）：固定范围的获客启动包，把老旧官网改造成可被搜索、可被 AI 理解、可承接咨询的本地获客入口，含上线与三个月本地 SEO 内容验证。 Local Growth Launch · 本地获客启动包 | BEE Sigma BEESigma 工作流 脱敏案例 看演示 博客 负责任 AI 市场雷达AIM 评估English EN 市场雷达 AIM 评估 ✦ 免费 AIV 诊断 GTM 工作流 · LOCAL GROWTH LAUNCH ## 官网应该带来本地客户，而不是躺在网上。 面向本地服务型中小企业的获客启动包，全球远程交付、以英语市场为主：一次固定范围的关键改造——让客户能找到你、看懂你、愿意联系你，随后用三个月内容验证效果。 全球远程交付 · 英语市场为主固定范围 · 一口价上线 + 3 个月内容验证 获取免费 AIV 初诊直接咨询 → WHY · 你的官网正在丢客户 ## 客户不是没需求，是找不到你 01 ## 搜不到 客户搜 “附近的汽修”“你所在城市的物业管理” 时你不在结果里，生意去了出现在第一屏的同行。 02 ## 看不懂 官网没说清你服务谁、覆盖哪些区域、凭什么可信，客户 10 秒内无法判断 “就是这家”。 03 ## 联系不上 没有明确的电话、表单或预约入口，看完页面的客户找不到下一步动作。 04 ## AI 在推荐别人 越来越多客户直接问 ChatGPT 和 Google AI “附近哪家靠谱”。AI 读不懂你的网站，就只会推荐竞争对手。 SCOPE · 启动包的边界 ## 一口价，边界清楚，不现场加戏 你付的每一分钱都对应下面的清单；清单之外的需求先报价、你确认后才做。 启动包包含 诊断AIV 初诊 + 现有网站与转化入口检查 + 同类客户对比，交付一份你能读懂的问题与优先级摘要。 网站重构最多 5 个标准页面：首页、服务页、关于我们、案例/评价页、联系/预约页。 本地 SEO 基础Title、Meta、服务区域、FAQ、基础 Schema、地图与联系入口——搜索引擎和 AI 都能识别你是谁、服务哪里。 转化设计电话、表单、预约或报价咨询 CTA，按你现有工具接入；客户能在页面完成一次明确的咨询动作。 三个月内容验证每月 2 篇本地搜索内容（发布前由你确认）+ 每月一次简报，为 90 天的升级判断留下依据。 明确不包含 电商、会员、复杂预约系统、CRM / ERP 接口与定制应用开发。 无限制页面、无限次视觉修改、品牌重塑、拍摄、广告投放及第三方软件费用。 Google 排名、流量、询盘量或成交量保证——我们只承诺把该做对的事做对、可验收。 范围之外的需求（更多页面、拍摄、投放等）以增购形式先报价、确认后执行。 PATH · 四步走 ## 从免费初诊到长期增长 每一步都以上一步的真实结果为前提——验证有效才继续，随时可以停。 01免费 AIV 初诊看清官网为什么没被搜索和 AI 推荐 02重构上线核心服务、区域与咨询入口清楚可见 03三个月内容验证用询盘与内容表现验证有效性 04升级获客工作流有效再进入 GTM、CRM 或 Golden Flow 定制流 FOR · 谁最适合 ## 为本地服务行业量身设计 REAL ESTATE & PROPERTY ## 地产与物业管理 改造前官网只有一页公司简介，业主搜 “property management + 城市名” 找到的全是同行。 改造后服务区域、管理范围、业主评价与预约入口清晰上线，AI 能准确介绍你管什么、在哪里管。 AUTO REPAIR ## 汽修 改造前客户来电反复问 “修不修变速箱、大概多少钱”，每天重复回答同样的问题。 改造后服务项目、价格区间与在线预约上线，Google 地图与官网互相印证，新客直接留单。 RENOVATION & HOME SERVICES ## 装修与家居服务 改造前案例照片全在朋友圈和微信里，说英语的客户完全看不到你的作品。 改造后双语案例页 + 报价咨询表单，本地搜索和 AI 推荐都能引用你的项目。 PREP · 你只需要提交这些 ## 一份在线资料，剩下交给我们 不开会、不排期。资料齐了就开工，初稿完成后发你在线确认。 01 ## 企业信息 企业名称、所属行业、服务区域（城市 / 街区）、营业时间。 02 ## 联系方式 电话、邮箱、WhatsApp / 微信，以及你希望客户用哪种方式咨询：电话、表单或预约链接。 03 ## 服务清单 3–8 项服务，每项一句话说明；有价格区间更好，没有也行。 04 ## 简介与资质 两三句话介绍你们是谁、做了几年；执照、认证或保险拍照发来即可。 05 ## 案例照片 3–5 张项目或门店照片，手机拍的就行；有前后对比更好。 06 ## 评价与 Logo 2–3 条客户评价（Google 评论截图即可）；Logo 有就发，没有我们用字标帮你做一个。 已有域名或旧网站？提供域名与后台权限即可；没有的我们帮你注册。 WHY BEE SIGMA · 不止建站 ## 网站只是起点，后面有路 我们不是建站公司。Local Growth Launch 是 BEE Sigma GTM 工作流的第一步——上线之后，AIV 持续诊断、市场雷达、内容官与线索评分随时可以接上，而不是交付后失联。 免费 AIV 初诊 负责指出问题，让你先看清值不值得修。 Local Growth Launch 负责把最关键的问题改掉，交付可验收的获客入口。 长期 Agent 工作流 负责持续对业务结果负责——只在验证有效后才建议进入。 我们不建议你一开始就花大钱做 AI 或复杂系统。先用一笔固定的小预算，把网站这个最基础的获客入口跑通：客户能找到你、看懂你、愿意联系你。三个月后，如果询盘和内容验证有效，再决定是否把获客、跟单和客户管理接入 Agent 工作流。 FAQ · 决定之前 ## 你可能想问的六个问题 ## 启动包都包含什么？怎么收费？ 范围是固定的：诊断、最多 5 页的网站重构、本地 SEO 基础、转化设计，以及三个月内容验证（每月 2 篇内容 + 一次简报）。一口价，正式报价在免费 AIV 初诊之后给出；范围之外的需求——更多页面、拍摄、广告投放、第三方软件费用——会先报价，你确认后才执行，不会现场加价。 ## 多久能上线？ 不需要开会或访谈。你在线提交一份资料（企业信息、联系方式、服务清单、简介与资质、案例照片、评价与 Logo——见上方清单），我们完成重构初稿后发你在线确认；已有域名或旧网站的提供后台权限，没有的我们帮你注册。资料齐、确认及时，上线就快。 ## 你们保证 Google 排名吗？ 不保证排名、流量或询盘量——任何敢做这种保证的服务商都值得警惕。我们承诺的是可验收的交付：搜索引擎和 AI 能正确识别你的企业、服务与区域；页面有明确的咨询动作；三个月内容按计划发布并复盘。 ## 我已经有网站了，能在原来的基础上改吗？ 可以。初诊会先检查现有网站，能保留的保留。重构的标准是 “客户 10 秒内看懂你是谁、服务哪里、怎么联系”，而不是为了重做而重做。 ## 三个月之后呢？ 第 90 天你会拿到一页升级判断：继续内容运营、进入 GTM / CRM 工作流或 Golden Flow 定制工作流，或者停止投入。用真实询盘和数据说话，不绑年费、不强制续约。 ## 为什么要先做免费 AIV 初诊？ 先让你看清问题，再决定花不花钱。初诊报告会指出官网为什么没被搜索和 AI 推荐；你认可这份问题清单，再决定是否启动 Local Growth Launch。 ## 把第一条业务流，交给 Agent 试试 先做一次免费 AIV 初诊，看看你的官网在 Google 和 AI 眼里长什么样——值不值得修，你自己判断。 开始免费 AIV 诊断 AIM 落地评估 → 直接沟通 → 相关服务新西兰 AI 技术实施机构新西兰 AI 咨询服务推荐新西兰企业流程自动化解决方案新西兰 AI 技术落地实践案例新西兰企业智能化转型服务新西兰企业管理软件比较新西兰 AI 数字员工新西兰建筑公司 AI 落地企业如何利用 AI 提升效率如何引入 AI 到我的业务中 © 2026 BEEBEE AI Limited · 新西兰注册运营 · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · 邮件联系 · Google 商家档案 首页 · 博客 · 负责任 AI ===== # Local Growth Launch · Website Rebuild + Local SEO for Service Businesses | BEE Sigma URL: https://beesigma.com/local-growth-launch-en Summary: For local service businesses worldwide (English-speaking markets first): a fixed-scope launch pack that turns an outdated website into a local lead engine… Local Growth Launch · Website Rebuild + Local SEO for Service Businesses | BEE Sigma BEESigma Workflows Casebook Demos Blog Responsible AI BizRadarAIM Assessment中文 中文 BizRadar AIM Assessment ✦ Free AIV Scan GTM WORKFLOW · LOCAL GROWTH LAUNCH ## Your website should bring local customers — not just sit online. Website Rebuild + Local SEO Launch for service businesses — delivered remotely, worldwide. One fixed-scope engagement covering website launch and 3 months of SEO content, so customers can find you, understand you, and contact you. Serving clients worldwideFixed scope · one pricelaunch + 3 months of content Get a Free AIV PreviewTalk to us → WHY · Your website is losing customers ## Demand exists — customers just can’t find you 01 ## Not found When customers search “auto repair near me” or “property management in your city”, you’re not in the results. The business goes to whoever is. 02 ## Not understood Your site doesn’t say who you serve, which areas you cover, or why you’re credible — visitors can’t tell “this is the one” within 10 seconds. 03 ## No next step No clear phone, form or booking entry. Customers finish the page and have nothing to click. 04 ## AI recommends others More customers now ask ChatGPT and Google AI “who’s good near me”. If AI can’t read your site, it recommends your competitors. SCOPE · Where the launch pack ends ## One price, clear boundaries, no surprise upsells Every dollar maps to the list below; anything beyond it is quoted first and only done after you approve. THE LAUNCH PACK INCLUDES DiagnosisAIV preview + audit of your current site and enquiry paths + peer comparison — delivered as a readable summary of issues and priorities. Website rebuildUp to 5 standard pages: Home, Services, About, Cases & Reviews, Contact & Booking. Local SEO foundationTitles, metas, service areas, FAQ, base schema, map and contact entries — so search engines and AI can tell who you are and where you serve. Conversion designPhone, form, booking or quote CTAs wired into the tools you already use; visitors can complete one clear enquiry action. 3 months of content2 local-search articles per month (you approve before publishing) + a monthly report, building the evidence for the day-90 decision. NOT INCLUDED E-commerce, memberships, complex booking systems, CRM / ERP integrations or custom app development. Unlimited pages, unlimited visual revisions, rebranding, photography, ad spend or third-party software costs. Guarantees of Google rankings, traffic, enquiries or revenue — we commit to doing the right things, verifiably. Needs beyond scope (extra pages, photography, ads) are quoted as add-ons and executed only after approval. PATH · Four steps ## From a free preview to long-term growth Each step builds on the real results of the last — continue only when it works, stop anytime. 01Free AIV previewsee why search and AI aren’t recommending you 02Rebuild & launchservices, areas and enquiry entries clearly visible 033-month validationverify with real enquiries and content performance 04Upgrade the funnelonly then move into GTM, CRM or a custom Golden Flow FOR · Built for local services ## Designed for local service businesses REAL ESTATE & PROPERTY ## Real estate & property management BEFOREA one-page company intro; owners searching “property management + your city” only find your competitors. AFTERService areas, management scope, owner reviews and booking entries live — AI can accurately describe what you manage and where. AUTO REPAIR ## Auto repair BEFORECustomers call to ask “do you do transmissions, roughly how much” — the same questions answered all day. AFTERServices, price ranges and online booking live; Google Maps and your website confirm each other, and new customers leave their details. RENOVATION & HOME SERVICES ## Renovation & home services BEFOREYour portfolio lives in scattered social posts — customers searching locally never see your work. AFTERA bilingual case page + quote form; local search and AI recommendations can both cite your projects. PREP · What you submit ## One online form — we do the rest No meetings, no scheduling. Once your materials are in, we start; you review the draft online. 01 ## Business basics Business name, industry, service areas (city / suburbs), opening hours. 02 ## Contact & enquiry Phone, email, WhatsApp / WeChat — and how you want customers to reach you: call, form, or a booking link. 03 ## Service list 3–8 services, one line each; price ranges help but aren’t required. 04 ## Intro & credentials Two or three sentences on who you are and how long you’ve operated; snap photos of licences, certifications or insurance. 05 ## Case photos 3–5 photos of projects or your premises — phone shots are fine; before/after pairs are even better. 06 ## Reviews & logo 2–3 customer reviews (Google review screenshots work); send a logo if you have one — if not, we’ll set a clean wordmark. Already have a domain or an old website? Just share the domain and backend access; if not, we’ll register one for you. WHY BEE SIGMA · More than a website ## The website is the start — there’s a road after it We’re not a web design shop. Local Growth Launch is step one of BEE Sigma’s GTM workflow — after launch, AIV monitoring, Market Radar, the Content Agent and lead scoring can plug in whenever you’re ready. No post-delivery silence. The free AIV preview points out the problems, so you see what’s worth fixing before spending. Local Growth Launch fixes the most critical ones and delivers a verifiable lead-capture entry. Long-term agent workflows stay accountable for business results — recommended only after validation. We don’t recommend starting with a big AI budget or complex systems. Start with a small, fixed budget and get the basics working: customers can find you, understand you, and want to contact you. After three months, if enquiries and content validate, then decide whether to plug lead-gen, follow-up and client management into agent workflows. FAQ · Before you decide ## Six questions worth asking ## What does the launch pack include? How is it priced? The scope is fixed: diagnosis, a rebuild of up to 5 pages, the local SEO foundation, conversion design, and 3 months of content validation (2 articles per month + a monthly report). One fixed price, formally quoted after the free AIV preview. Anything beyond — extra pages, photography, ad spend, third-party software — is quoted first and only executed after your approval. No on-the-spot upsells. ## How long until launch? No meetings or interviews. You submit one set of materials online — business basics, contact details, service list, intro and credentials, case photos, reviews and logo (see the checklist above). We build the rebuild draft and send it for your online review; if you have a domain or an old site, share the access — if not, we register one for you. Complete materials and prompt confirmations make launches fast. ## Do you guarantee Google rankings? No — and be wary of anyone who does. What we commit to is verifiable delivery: search engines and AI correctly identify your business, services and areas; the pages carry one clear enquiry action; and three months of content is published and reviewed on schedule. ## I already have a website. Can you build on it? Yes. The diagnosis reviews your current site first and keeps what works. The rebuild standard is “a customer understands who you are, where you serve and how to contact you within 10 seconds” — not rebuilding for its own sake. ## What happens after three months? On day 90 you receive a one-page decision: continue content operations, move into GTM / CRM workflows or a custom Golden Flow, or stop investing. Real enquiries and data decide — no annual lock-in, no forced renewal. ## Why start with the free AIV preview? So you see the problems before spending anything. The preview report shows why search and AI aren’t recommending your site; if you agree with the issue list, then decide whether to start Local Growth Launch. ## Hand your first workflow to the agents Start with a free AIV preview — see how your website looks to Google and AI, and judge for yourself whether it’s worth fixing. Start the Free AIV Scan AIM Assessment → Contact us → RelatedAI Implementation Partners NZAI Consulting AucklandWorkflow Automation SolutionsAI Implementation Case StudiesAI Transformation Services NZNZ Business Software ComparedAI Digital Employees NZAI for Construction Companies NZHow Businesses Use AI to Improve EfficiencyHow to Introduce AI into Your Business © 2026 BEEBEE AI Limited · Registered in New Zealand · NZBN 9429051781781 · B:Hive, 74 Taharoto Road, Smales Farm, Takapuna, Auckland · Email us · Google Business Profile Home · Blog · Responsible AI